Skip to left side bar
>
  • File
  • Edit
  • View
  • Run
  • Kernel
  • Tabs
  • Settings
  • Help

Open Tabs

  • 08-visualization-plotly.ipynb
  • 09-visualization-seaborn.ipynb
  • 10-databases-sql.ipynb
  • 11-databases-mongodb.ipynb
  • 12-ml-core.ipynb
  • 13-ml-data-pre-processing-and-production.ipynb
  • 14-ml-classification.ipynb
  • 15-ml-regression.ipynb

Kernels

  • 10-databases-sql.ipynb
  • 11-databases-mongodb.ipynb
  • 13-ml-data-pre-processing-and-production.ipynb
  • 12-ml-core.ipynb
  • 14-ml-classification.ipynb
  • 15-ml-regression.ipynb
  • 09-visualization-seaborn.ipynb
  • 08-visualization-plotly.ipynb

Terminals

    //ds-curriculum/@textbook/
    Name
    ...
    Last Modified
    • .ipynb_checkpoints20 hours ago
    • data2 months ago
    • 01-python-getting-started.2022-12-23T07-21-48-609Z.ipynba day ago
    • 01-python-getting-started.ipynb2 months ago
    • 02-python-advanced.ipynba day ago
    • 03-pandas-getting-started.2022-12-23T07-21-48-609Z.ipynb2 months ago
    • 03-pandas-getting-started.ipynb2 months ago
    • 04-pandas-advanced.2022-12-23T07-21-48-609Z.ipynba day ago
    • 04-pandas-advanced.ipynb2 months ago
    • 05-pandas-summary-statistics.2022-12-23T07-21-48-609Z.ipynba day ago
    • 05-pandas-summary-statistics.ipynb2 months ago
    • 06-visualization-matplotlib.2022-12-23T07-21-48-609Z.ipynb21 hours ago
    • 06-visualization-matplotlib.ipynb2 months ago
    • 07-visualization-pandas.ipynb20 hours ago
    • 08-visualization-plotly.ipynb20 hours ago
    • 09-visualization-seaborn.ipynb20 hours ago
    • 10-databases-sql.ipynb2 months ago
    • 11-databases-mongodb.ipynb20 hours ago
    • 12-ml-core.ipynb2 months ago
    • 13-ml-data-pre-processing-and-production.ipynb20 hours ago
    • 14-ml-classification.ipynb20 hours ago
    • 15-ml-regression.ipynba month ago
    • 16-ml-unsupervised-learning.ipynb2 months ago
    • 17-ts-core.ipynb2 months ago
    • 18-ts-models.ipynb2 months ago
    • 19-linux-command-line.ipynb2 months ago
    • 20-statistics.ipynb2 months ago
    • 21-python-object-oriented-programming.ipynb2 months ago
    • 22-apis.ipynb2 months ago
    • main.py3 months ago
    ​x
     

    Usage Guidelines

    This lesson is part of the DS Lab core curriculum. For that reason, this notebook can only be used on your WQU virtual machine.

    This means:

    • ⓧ No downloading this notebook.
    • ⓧ No re-sharing of this notebook with friends or colleagues.
    • ⓧ No downloading the embedded videos in this notebook.
    • ⓧ No re-sharing embedded videos with friends or colleagues.
    • ⓧ No adding this notebook to public or private repositories.
    • ⓧ No uploading this notebook (or screenshots of it) to other websites, including websites for study resources.

    <font size="+3"><strong>Databases: SQL</strong></font>

    Databases: SQL

    [ ]:
    from IPython.display import YouTubeVideo
    # Working with SQL Databases

    Working with SQL Databases¶

    A database is a collection of interrelated data. The primary goal of a database is to store and retrieve information in a convenient and efficient way. There are many types of databases. In this section, we will be dealing with a **relational database**. A relational database is a widely used database model that consists of a collection of uniquely named **tables** used to store information. The structure of a database model with its tables, constraints, and relationships is called a **schema**. 

    A database is a collection of interrelated data. The primary goal of a database is to store and retrieve information in a convenient and efficient way. There are many types of databases. In this section, we will be dealing with a relational database. A relational database is a widely used database model that consists of a collection of uniquely named tables used to store information. The structure of a database model with its tables, constraints, and relationships is called a schema.

    A Structured Query Language (SQL), is used to retrieve information from a relational database. SQL is one of the most commonly used database languages. It allows data stored in a relational database to be queried, modified, and manipulated easily with basic commands. SQL powers database engines like MySQL, SQL Server, SQLite, and PostgreSQL. The examples and projects in this course will use SQLite.

    A table refers to a collection of rows and columns in a relational database. When reading data into a pandas DataFrame, an index can be defined, which acts as the label for every row in the DataFrame.

    # Connecting to a Database

    Connecting to a Database¶

    ## ipython-sql 

    ipython-sql¶

    ### Magic Commands

    Magic Commands¶

    Jupyter notebooks can run code that is not valid Python code but still affect the notebook . These special commands are called magic commands. Magic commands can have a range of properties. Some commonly used magic functions are below:

    Jupyter notebooks can run code that is not valid Python code but still affect the notebook . These special commands are called magic commands. Magic commands can have a range of properties. Some commonly used magic functions are below:

    Magic Command Description of Command
    %pwd Print the current working directory
    %cd Change the current working directory
    %ls List the contents of the current directory
    %history Show the history of the In [ ]: commands

    We will be leveraging magic commands to work with a SQLite database.

    ### ipython-sql

    ipython-sql¶

    `ipython-sql` allows you to write SQL code directly in a Jupyter Notebook. The `%sql` (or `%%sql`) magic command is added to the beginning of a code block and then SQL code can be written.

    ipython-sql allows you to write SQL code directly in a Jupyter Notebook. The %sql (or %%sql) magic command is added to the beginning of a code block and then SQL code can be written.

    ### Connecting with ipython-sql

    Connecting with ipython-sql¶

    We can connect to a database using the %sql magic function:

    We can connect to a database using the %sql magic function:

    [ ]:
    %sql sqlite:////home/jovyan/nepal.sqlite
    ## sqlite3

    sqlite3¶

    We can also connect to the same database using the sqlite3 package:

    We can also connect to the same database using the sqlite3 package:

    [ ]:
    conn = sqlite3.connect("/home/jovyan/nepal.sqlite")
    # Querying a Database

    Querying a Database¶

    ## Building Blocks of the Basic Query

    Building Blocks of the Basic Query¶

    There are six common clauses used for querying data:

    There are six common clauses used for querying data:

    Clause Name Definition
    SELECT Determines which columns to include in the query's result
    FROM Identifies the table from which to query the data from
    WHERE filters data
    GROUP BY groups rows by common values in columns
    HAVING filters out unwanted groups from GROUP BY
    ORDER BY Orders the rows using one or more columns
    LIMIT Outputs the specified number of rows

    All clauses may be used together, but SELECT and FROM are the only required clauses. The format of clauses is in the example query below:

    SELECT column1, column2
    FROM table_name
    WHERE "conditions"
    GROUP BY "column-list"
    HAVING "conditions"
    ORDER BY "column-list"
    
    ## SELECT and FROM

    SELECT and FROM¶

    You can use `SELECT *` to select all columns in a table. `FROM` specifies the table in the database to query. `LIMIT 5` will select only the first five rows. 

    You can use SELECT * to select all columns in a table. FROM specifies the table in the database to query. LIMIT 5 will select only the first five rows.

    Example

    [ ]:
    FROM id_map
    You can also use `SELECT` to select certain columns in a table

    You can also use SELECT to select certain columns in a table

    [ ]:
    SELECT household_id,
    <font size="+1">Practice</font>

    Practice

    Try it yourself! Use SELECT to select the district_id column from the id_map table.

    [ ]:
    %%sql
    We can also assign an **alias** or temporary name to a column using the `AS` command. Aliases can also be used on a table. See the example below, which assigns the alias `household_number` to `household_id`

    We can also assign an alias or temporary name to a column using the AS command. Aliases can also be used on a table. See the example below, which assigns the alias household_number to household_id

    [ ]:
    SELECT household_id AS household_number,
    <font size="+1">Practice</font>

    Practice

    Try it yourself! Use SELECT, FROM, AS, and LIMIT to select the first 5 rows from the id_map table. Rename the district_id column to district_number.

    [ ]:
    %%sql
    ## Filtering and Sorting Data

    Filtering and Sorting Data¶

    SQL provides a variety of comparison operators that can be used with the WHERE clause to filter the data. 

    SQL provides a variety of comparison operators that can be used with the WHERE clause to filter the data.

    Comparison Operator Description
    = Equal
    > Greater than
    < Less than
    >= Greater than or equal to
    <= Less than or equal to
    <> or != Not equal to
    LIKE String comparison test
    For example, to select the first 5 homes in Ramechhap (district `2`):

    For example, to select the first 5 homes in Ramechhap (district 2):

    [ ]:
    %%sql
    <font size="+1">Practice</font>

    Practice

    Try it yourself! Use WHERE to select the row with household_id equal to 13735001

    [ ]:
    %%sql
    ## Aggregating Data

    Aggregating Data¶

    Aggregation functions take a collection of values as inputs and return one value as the output. The table below gives the frequently used built-in aggregation functions:

    Aggregation functions take a collection of values as inputs and return one value as the output. The table below gives the frequently used built-in aggregation functions:

    Aggregation Function Definition
    MIN Return the minimum value
    MAX Return the largest value
    SUM Return the sum of values
    AVG Return the average of values
    COUNT Return the number of observations
    Use the `COUNT` function to find the number of observations in the `id_map` table that come from Ramechhap (district `2`):

    Use the COUNT function to find the number of observations in the id_map table that come from Ramechhap (district 2):

    [ ]:
    WHERE district_id = 2
    Aggregation functions are frequently used with a `GROUP BY` clause to perform the aggregation on groups of data. For example, the query below returns the count of observations in each District:

    Aggregation functions are frequently used with a GROUP BY clause to perform the aggregation on groups of data. For example, the query below returns the count of observations in each District:

    [ ]:
    GROUP BY district_id
     `DISTINCT` is a keyword to select unique records in a query result. For example, if we want to know the unique values in the `district_id` column:

    DISTINCT is a keyword to select unique records in a query result. For example, if we want to know the unique values in the district_id column:

    [ ]:
    SELECT distinct(district_id)
    <font size="+1">Practice</font>

    Practice

    Try it yourself! Use DISTINCT to count the number of unique values in the vdcmun_id column.

    [ ]:
    %%sql
    `DISTINCT` and `COUNT` can be used in combination to count the number of distinct records. For example, if we want to know the number of unique values in the `district_id` column:

    DISTINCT and COUNT can be used in combination to count the number of distinct records. For example, if we want to know the number of unique values in the district_id column:

    [ ]:
    SELECT count(distinct(district_id))
    <font size="+1">Practice</font>

    Practice

    Try it yourself! Use DISTINCT and COUNT to count the number of unique values in the vdcmun_id column.

    [ ]:
    %%sql
    # Joining Tables

    Joining Tables¶

    Joins link data from two or more tables together by using a column that is common between the two tables. The basic syntax for a join is below, where `table1` and `table2` refer to the two tables being joined, `column1` and `column2` refer to columns to be returned from both tables, and `ID` refers to the common column in the two tables. 

    Joins link data from two or more tables together by using a column that is common between the two tables. The basic syntax for a join is below, where table1 and table2 refer to the two tables being joined, column1 and column2 refer to columns to be returned from both tables, and ID refers to the common column in the two tables.

    SELECT table1.column1,
           table2.column2
    FROM table_1
    JOIN table2 ON table1.id = table1.id
    
    We'll explore the concept of joins by first identifying a single household that we'd like to pull in building information for. For example, let's say we want to see the corresponding `foundation_type` for the first home in Ramechhap (District 1). We'll start by looking at this single record in the `id_map` table.

    We'll explore the concept of joins by first identifying a single household that we'd like to pull in building information for. For example, let's say we want to see the corresponding foundation_type for the first home in Ramechhap (District 1). We'll start by looking at this single record in the id_map table.

    [ ]:
    WHERE district_id = 2
    This household has `building_id` equal to 23. Let's look at the `foundation_type` for this building, by filtering the `building_structure` table to find this building.

    This household has building_id equal to 23. Let's look at the foundation_type for this building, by filtering the building_structure table to find this building.

    [ ]:
    FROM building_structure
    To join the two tables and limit the results to `building_id = 23`:    

    To join the two tables and limit the results to building_id = 23:

    [ ]:
    JOIN building_structure ON id_map.building_id = building_structure.building_id
    In addition to the basic `JOIN` clause, specific join types can be specified, which specify whether the common column needs to be in one, both, or either of the two tables being joined. The different join types are below. The left table is the table specified first, immediately after the `FROM` clause and the right table is the table specified after the `JOIN` clause. If the generic `JOIN` clause is used, then by default the `INNER JOIN` will be used.

    In addition to the basic JOIN clause, specific join types can be specified, which specify whether the common column needs to be in one, both, or either of the two tables being joined. The different join types are below. The left table is the table specified first, immediately after the FROM clause and the right table is the table specified after the JOIN clause. If the generic JOIN clause is used, then by default the INNER JOIN will be used.

    JOIN Type Definition
    INNER JOIN Returns rows where ID is in both tables
    LEFT JOIN Returns rows where ID is in the left table. Return NA for values in column, if ID is not in right table.
    RIGHT JOIN Returns rows where ID is in the right table. Return NA for values in column, if ID is not in left table.
    FULL JOIN Returns rows where ID is in either table. Return NA for values in column, if ID is not in either table.
    WQU WorldQuant University Applied Data Science Lab QQQQ
    The video below outlines the main types of joins:

    The video below outlines the main types of joins:

    [ ]:
    YouTubeVideo("2HVMiPPuPIM")
    <font size="+1">Practice</font>

    Practice

    Try it yourself! Use the DISTINCT command to create a column with all unique building IDs in the id_map table. LEFT JOIN this column with the roof_type column from the building_structure table, showing only buildings where district_id is 1 and limiting your results to the first five rows of the new table.

    [ ]:
    %%sql
    # Using pandas with SQL Databases

    Using pandas with SQL Databases¶

    To save the output of a query into a pandas DataFrame, we will use connect to the SQLite database using the SQLite3 package:

    To save the output of a query into a pandas DataFrame, we will use connect to the SQLite database using the SQLite3 package:

    [ ]:
    conn = sqlite3.connect("/home/jovyan/nepal.sqlite")
    To run a query using `sqlite3`, we need to store the query as a string. For example, the variable below called `query` is a string containing a query which returns the first 10 rows from the `id_map` table:

    To run a query using sqlite3, we need to store the query as a string. For example, the variable below called query is a string containing a query which returns the first 10 rows from the id_map table:

    [ ]:
        FROM id_map
    To save the results of the query into a pandas DataFrame, use the `pd.read_sql()` function. The optional parameter `index_col` can be used to set the index to a specific column from the query. 

    To save the results of the query into a pandas DataFrame, use the pd.read_sql() function. The optional parameter index_col can be used to set the index to a specific column from the query.

    [ ]:
    df = pd.read_sql(query, conn, index_col="building_id")
    <font size="+1">Practice</font>

    Practice

    Try it yourself! Use the pd.read_sql function to save the results of a query to a DataFrame. The query should select first 20 rows from the id_map table.

    [ ]:
    query = ...
    # References & Further Reading

    References & Further Reading¶

    • Additional Explanation of Magic Commands
    • ipython-SQL User Documentation
    • Data Carpentry Course on SQL in Python
    • SQL Course Material on GitHub (1)
    • SQL Course Material on GitHub (2)
    ---

    Copyright 2022 WorldQuant University. This content is licensed solely for personal use. Redistribution or publication of this material is strictly prohibited.

    ​x
     

    Usage Guidelines

    This lesson is part of the DS Lab core curriculum. For that reason, this notebook can only be used on your WQU virtual machine.

    This means:

    • ⓧ No downloading this notebook.
    • ⓧ No re-sharing of this notebook with friends or colleagues.
    • ⓧ No downloading the embedded videos in this notebook.
    • ⓧ No re-sharing embedded videos with friends or colleagues.
    • ⓧ No adding this notebook to public or private repositories.
    • ⓧ No uploading this notebook (or screenshots of it) to other websites, including websites for study resources.

    <font size="+3"><strong>Machine Learning: Linear Regression</strong></font>

    Machine Learning: Linear Regression

    # Linear Regression

    Linear Regression¶

    [1]:
    xxxxxxxxxx
     
    from IPython.display import YouTubeVideo
    In machine learning, a **regression** problem is when you need to build a model that's going to predict a continuous, numerical value, like the sale price of an apartment. One of the models that you can use for regression problems is called **linear regression**. In it's simplest form, we fit a model that will predict a single output variable (called a **target vector**) as a linear function of a single input variable (called a **feature matrix**). 

    In machine learning, a regression problem is when you need to build a model that's going to predict a continuous, numerical value, like the sale price of an apartment. One of the models that you can use for regression problems is called linear regression. In it's simplest form, we fit a model that will predict a single output variable (called a target vector) as a linear function of a single input variable (called a feature matrix).

    Speaking mathematically, if we have input data points 𝑥x and corresponding measured output 𝑦y, then we find parameters 𝑚m and 𝑏b such that 𝑦≈𝑚×𝑥+𝑏y≈m×x+b for our measured data points. We then use the fitted values of 𝑚m and 𝑏b to predict values of 𝑦y for new values of 𝑥x.

    ## Fitting a Model to Training Data

    Fitting a Model to Training Data¶

    You'll work on two cases: a model on the raw data set and a model on transformed data. First try to use linear regression to predict `price_aprox_usd` as a multiple of `surface_covered_in_m2` and the addition of a constant for the `mexico-city-real-estate-1.csv` dataset.<span style='color: transparent; font-size:1%'>WQU WorldQuant University Applied Data Science Lab QQQQ</span>

    You'll work on two cases: a model on the raw data set and a model on transformed data. First try to use linear regression to predict price_aprox_usd as a multiple of surface_covered_in_m2 and the addition of a constant for the mexico-city-real-estate-1.csv dataset.WQU WorldQuant University Applied Data Science Lab QQQQ

    [2]:
    x
    import pandas as pd
    from sklearn.linear_model import LinearRegression
    ​
    # Import data
    columns = ["surface_covered_in_m2", "price_aprox_usd"]
    mexico_city1 = pd.read_csv("./data/mexico-city-real-estate-1.csv", usecols=columns)
    ​
    # Drop rows with missing values
    # (or you could use an imputer ☝️)
    mexico_city1.dropna(inplace=True)
    ​
    # Split data into feature matrix
    X = mexico_city1[["surface_covered_in_m2"]]
    y = mexico_city1["price_aprox_usd"]
    ​
    # Instantiate predictor
    lr = LinearRegression()
    ​
    # Fit predictor to data
    lr.fit(X, y)
    [2]:
    LinearRegression()
    In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
    On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
    LinearRegression()
    <font size="+1">Practice</font> 

    Practice

    Fit a linear regression model to the mexico-city-real-estate-2.csv data set to relate "price_aprox_usd" and "surface_covered_in_m2".

    [3]:
    xxxxxxxxxx
     
    # Import data
    columns = ["price_aprox_usd", "surface_covered_in_m2"]
    mexico_city2 = pd.read_csv("./data/mexico-city-real-estate-2.csv", usecols=columns)
    # Drop rows with missing values
    mexico_city2.dropna(inplace=True)
    ​
    # Split data into feature matrix
    X = mexico_city2[["surface_covered_in_m2"]]
    y = mexico_city2["price_aprox_usd"]
    ​
    # Instantiate predictor
    lr = LinearRegression()
    ​
    # Fit predictor to data
    lr.fit(X,y)
    ​
    [3]:
    LinearRegression()
    In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
    On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
    LinearRegression()
    ## Generating Predictions Using a Trained Model

    Generating Predictions Using a Trained Model¶

    After fitting the model, we want to use it to make predictions. In most applications, you'll want to predict an unknown quantity from data that's different from the data you've fitted our model on. To test the accuracy of your fitted model, you'll typically use a different set of data with an outcome you already know. Here, we'll use the dataset from `mexico-city-test-features.csv` and `mexico-city-test-labels.csv`.  It's also helpful to plot the data and predicted data to see if there are any patterns that suggest fitting a different model.

    After fitting the model, we want to use it to make predictions. In most applications, you'll want to predict an unknown quantity from data that's different from the data you've fitted our model on. To test the accuracy of your fitted model, you'll typically use a different set of data with an outcome you already know. Here, we'll use the dataset from mexico-city-test-features.csv and mexico-city-test-labels.csv. It's also helpful to plot the data and predicted data to see if there are any patterns that suggest fitting a different model.

    [4]:
    x
    # Import data
    mexico_city_features = pd.read_csv(
        "./data/mexico-city-test-features.csv", usecols=["surface_covered_in_m2"]
    )
    mexico_city_labels = pd.read_csv("./data/mexico-city-test-labels.csv")
    ​
    # Drop missing values
    mexico_city_features.dropna(inplace=True)
    ​
    # Generate predictions
    price_pred_example = lr.predict(mexico_city_features)
    ​
    # Print predictions
    price_pred_example[:5]
    [4]:
    array([309549.84644749, 309411.49120101, 310207.03386826, 309428.78560682,
           309515.25763587])
    <font size="+1">Practice</font> 

    Practice

    Read the data from mexico-city-real-estate-4.csv into a DataFrame and then generate a list of price predictions for the properties using your model lr.

    [5]:
    xxxxxxxxxx
     
    # Import data
    mexico_city4 = pd.read_csv( "./data/mexico-city-test-features.csv",usecols=["surface_covered_in_m2"])
    ​
    # Drop missing values
    mexico_city4.dropna(inplace=True)
    ​
    # Generate predictions
    price_pred = lr.predict(mexico_city4)
    ​
    # Print predictions
    price_pred[:5]
    [5]:
    array([309549.84644749, 309411.49120101, 310207.03386826, 309428.78560682,
           309515.25763587])
    ## Ridge Regression

    Ridge Regression¶

    Sometimes,the values for coefficients and the intercept - both positive and negative - are very large. When you see this in a linear model — especially a high-dimensional model — what's happening is that the model is overfitting to the training data and then can't generalize to the test data. Some people call this the curse of dimensionality. ☠️

    The way to solve this problem is to use regularization, a group of techniques that prevent overfitting. In this case, we'll change the predictor from LinearRegression to Ridge, which is a linear regressor with an added tool for keeping model coefficients from getting too big.

    Here's a good explanation of what a ridge regression is and why it's important:

    [6]:
    xxxxxxxxxx
     
    YouTubeVideo("Q81RR3yKn30")
    [6]:
    ## Generalization

    Generalization¶

    Notice that we tested the model with a dataset that's *different* from the one we used to train the model. Machine learning models are useful if they allow you to make predictions about data other than what you used to train your model. We call this concept **generalization**.  By testing your model with different data than you used to train it, you're checking to see if your model can generalize.  Most machine learning models do not generalize to all possible types of input data, so they should be used with care. On the other hand, machine learning models that don't generalize to make predictions for at least a restricted set of data aren't very useful.

    Notice that we tested the model with a dataset that's different from the one we used to train the model. Machine learning models are useful if they allow you to make predictions about data other than what you used to train your model. We call this concept generalization. By testing your model with different data than you used to train it, you're checking to see if your model can generalize. Most machine learning models do not generalize to all possible types of input data, so they should be used with care. On the other hand, machine learning models that don't generalize to make predictions for at least a restricted set of data aren't very useful.

    ## Calculating the Mean Absolute Error for a List of Predictions

    Calculating the Mean Absolute Error for a List of Predictions¶

    Plots are great for displaying information, but a value that tells you the typical error in a prediction is helpful too. This value is called the **mean absolute error**, and it's defined as the average value of the magnitude of the error in the predictions. The closer the MAE is to `0`, the better our model fits the data. scikit-learn will do this for you if you pass it the price predictions from your regression model and the actual prices from the test data set. Let's see how our `lr` model did by comparing its predictions to the true values in `mexico_city_labels`.

    Plots are great for displaying information, but a value that tells you the typical error in a prediction is helpful too. This value is called the mean absolute error, and it's defined as the average value of the magnitude of the error in the predictions. The closer the MAE is to 0, the better our model fits the data. scikit-learn will do this for you if you pass it the price predictions from your regression model and the actual prices from the test data set. Let's see how our lr model did by comparing its predictions to the true values in mexico_city_labels.

    [7]:
    xxxxxxxxxx
     
    from sklearn.metrics import mean_absolute_error
    ​
    mean_absolute_error(price_pred_example, mexico_city_labels)
    [7]:
    226209.01327442465
    ## Access an Attribute of a Trained Model

    Access an Attribute of a Trained Model¶

    After training a model that fits a straight line to your data, you can now obtain the parameters that fit your line. We're particularly interested in the slope `regr_lr.coef_` and the axis intercept `regr_lr.intercept_`

    After training a model that fits a straight line to your data, you can now obtain the parameters that fit your line. We're particularly interested in the slope regr_lr.coef_ and the axis intercept regr_lr.intercept_

    [8]:
    xxxxxxxxxx
     
    print(lr.coef_)
    [3.45888116]
    
    [9]:
    xxxxxxxxxx
     
    print(lr.intercept_)
    309238.5471429155
    
    ## Multicollinearity

    Multicollinearity¶

    When you're creating a linear model that uses many features to make predictions, some of those features can be highly correlated with each other. This isn't a problem that's going to break your model; it will still make predictions and it might have good performance metrics. But it is an issue if you want to interpret the coefficients for your model because it becomes hard to tell which features are truly important. 

    When you're creating a linear model that uses many features to make predictions, some of those features can be highly correlated with each other. This isn't a problem that's going to break your model; it will still make predictions and it might have good performance metrics. But it is an issue if you want to interpret the coefficients for your model because it becomes hard to tell which features are truly important.

    Let's look at mexico-city-real-estate-1.csv for an example. First we'll import the data.

    [10]:
    xxxxxxxxxx
     
    import pandas as pd
    from sklearn.linear_model import LinearRegression
    ​
    # Import data
    columns = [
        "price",
        "price_aprox_local_currency",
        "price_aprox_usd",
        "surface_total_in_m2",
        "surface_covered_in_m2",
        "price_per_m2",
    ]
    mexico_city1 = pd.read_csv("./data/mexico-city-real-estate-1.csv", usecols=columns)
    ​
    # Drop missing values
    mexico_city1.dropna(inplace=True)
    ​
    mexico_city1.head()
    [10]:
    price price_aprox_local_currency price_aprox_usd surface_total_in_m2 surface_covered_in_m2 price_per_m2
    2 2700000.0 2748947.10 146154.51 61.0 61.0 44262.295082
    3 6347000.0 6462061.92 343571.36 176.0 128.0 49585.937500
    4 6870000.0 6994543.16 371882.03 180.0 136.0 50514.705882
    5 6500000.0 6617835.61 351853.45 300.0 300.0 21666.666667
    6 670000.0 682146.11 36267.97 65.0 65.0 10307.692308
    Now let's find the correlations between the columns.

    Now let's find the correlations between the columns.

    [11]:
    xxxxxxxxxx
     
    mexico_city1.corr()
    [11]:
    price price_aprox_local_currency price_aprox_usd surface_total_in_m2 surface_covered_in_m2 price_per_m2
    price 1.000000 0.333655 0.333655 0.112588 0.371073 0.380879
    price_aprox_local_currency 0.333655 1.000000 1.000000 0.118123 0.598506 -0.068775
    price_aprox_usd 0.333655 1.000000 1.000000 0.118123 0.598506 -0.068775
    surface_total_in_m2 0.112588 0.118123 0.118123 1.000000 0.125032 0.003488
    surface_covered_in_m2 0.371073 0.598506 0.598506 0.125032 1.000000 -0.147158
    price_per_m2 0.380879 -0.068775 -0.068775 0.003488 -0.147158 1.000000
    Let's see what happens when we fit a linear regression model for `surface_covered_in_m2` as a function of `price_aprox_usd` and `price_aprox_local_currency`.

    Let's see what happens when we fit a linear regression model for surface_covered_in_m2 as a function of price_aprox_usd and price_aprox_local_currency.

    [12]:
    xxxxxxxxxx
     
    lr = LinearRegression()
    lr.fit(
        mexico_city1[["price_aprox_usd", "price_aprox_local_currency"]],
        mexico_city1["surface_covered_in_m2"],
    )
    [12]:
    LinearRegression()
    In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
    On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
    LinearRegression()
    Let's take a look at the coefficients of the model:

    Let's take a look at the coefficients of the model:

    [13]:
    xxxxxxxxxx
     
    print(lr.coef_)
    [6593.61513754 -350.56569568]
    
    Ask yourself: Does it make sense that increasing the price of a property by one US dollar would translate to a 6593 m<sup>2</sup> increase in size? Perhaps, though it seems unlikely. And does it make sense that increasing the price by one Mexican peso would translate to a 350 m<sup>2</sup> *decrease* in size? Definitely not. So while this model may perform well when we evaluate it using metrics like mean absolute error, we can't use it to determine which features actually our target.

    Ask yourself: Does it make sense that increasing the price of a property by one US dollar would translate to a 6593 m2 increase in size? Perhaps, though it seems unlikely. And does it make sense that increasing the price by one Mexican peso would translate to a 350 m2 decrease in size? Definitely not. So while this model may perform well when we evaluate it using metrics like mean absolute error, we can't use it to determine which features actually our target.

    *References & Further Reading*

    References & Further Reading

    • A primer on linear regression
    • More on resampling from the pandas documentation
    • More information on rolling averages
    • More on absolute and mean absolute errors
    • A discussion of the various uses of model fitting in machine learning
    • Wikipedia Page on Multicollinearity
    • Online Article on Multicollinearity
    • Wikipedia Article on Generalization
    • Online Tutorial on Regression with scikit-learn
    • Official scikit-learn Documentation on Linear Models
    • Wikipedia Article on Logarithm Function
    ---

    Copyright 2022 WorldQuant University. This content is licensed solely for personal use. Redistribution or publication of this material is strictly prohibited.

    ​x
     

    Usage Guidelines

    This lesson is part of the DS Lab core curriculum. For that reason, this notebook can only be used on your WQU virtual machine.

    This means:

    • ⓧ No downloading this notebook.
    • ⓧ No re-sharing of this notebook with friends or colleagues.
    • ⓧ No downloading the embedded videos in this notebook.
    • ⓧ No re-sharing embedded videos with friends or colleagues.
    • ⓧ No adding this notebook to public or private repositories.
    • ⓧ No uploading this notebook (or screenshots of it) to other websites, including websites for study resources.

    <font size="+3"><strong>Machine Learning: Data Pre-Processing and Production</strong></font>

    Machine Learning: Data Pre-Processing and Production

    [1]:
    warnings.simplefilter(action="ignore", category=FutureWarning)
    # What's scikit-learn?

    What's scikit-learn?¶

    scikit-learn is a Python library that contains implementations of many common machine learning algorithms and uses common interfaces for these that enables experimentation. In this section, we'll look at linear regression (which you'll use to predict price based on the area of a property) and K-nearest neighbors, which you'll use to classify the neighborhood a property is in.

    # Data Preprocessing

    Data Preprocessing¶

    # Standardization

    Standardization¶

    **Standardization** is a widely used scaling technique to transform features before fitting into models. Feature scaling changes all a dataset's continuous features to give us a more consistent range of values. Specifically, we subtract the mean from each data point and then divide by the standard deviation:

    Standardization is a widely used scaling technique to transform features before fitting into models. Feature scaling changes all a dataset's continuous features to give us a more consistent range of values. Specifically, we subtract the mean from each data point and then divide by the standard deviation:

    𝑋̂ =𝑋−𝜇𝜎,X^=X−μσ,

    The goal of standardization is to improve model performance having all continuous features be on the same scale. It's useful in at least two circumstances:

    1. For machine leaning algorithms that use Euclidean distance (k-means and k-nearest neighbors), different scales can distort the calculation of distance and hurt model performance.
    2. For dimensionality reduction (principal component analysis), it can improve the model's ability to finds combinations of features that have the most variance.
    Let's check the following example where we apply standardization on one of the columns in the following DataFrame:

    Let's check the following example where we apply standardization on one of the columns in the following DataFrame:

    [2]:
    df = pd.read_csv("./data/mexico-city-test-features.csv").dropna()
    [2]:
    surface_covered_in_m2 lat lon neighborhood
    0 90.0 19.367931 -99.170262 Benito Juárez
    1 50.0 19.363542 -99.224084 Álvaro Obregón
    2 280.0 19.457982 -99.192690 Miguel Hidalgo
    3 55.0 19.334270 -99.083374 Iztapalapa
    4 80.0 19.416881 -99.109781 Venustiano Carranza
    Our target feature is the `"surface_covered_in_m2"` column. Let's first check the maximum and minimum of this column before standardization:

    Our target feature is the "surface_covered_in_m2" column. Let's first check the maximum and minimum of this column before standardization:

    [3]:
    print("Maximum before standardization is:", df["surface_covered_in_m2"].max())
    Maximum before standardization is: 280.0
    Minimum before standardization is: 50.0
    
    We can perform the transformation by first instantiating the scaler and assigning the feature to a variable name. Then we fit the scaler and transform the data:

    We can perform the transformation by first instantiating the scaler and assigning the feature to a variable name. Then we fit the scaler and transform the data:

    [4]:
    from sklearn.preprocessing import StandardScaler
    [5]:
    # Fit the scaler to feature
    [5]:
    StandardScaler()
    In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
    On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
    StandardScaler()
    [6]:
    # Pass the scaler to feature to transform data
    [6]:
    array([[ 1.62304525e-01],
           [-1.11902808e+00],
           [ 6.24863439e+00],
           ...,
           [ 2.13794980e-03],
           [-3.18195201e-01],
           [ 2.13794980e-03]])
    Now you can see the transformed data range is much smaller after standardization:

    Now you can see the transformed data range is much smaller after standardization:

    [7]:
    print("Maximum after standardization is:", X_transformed.max())
    Maximum after standardization is: 6.248634385593622
    Minimum after standardization is: -1.1190280771385155
    
    We can also combine the fit and transform process together into one step:

    We can also combine the fit and transform process together into one step:

    [8]:
    X_transformed = scaler.fit_transform(X_train)
    [8]:
    array([[ 1.62304525e-01],
           [-1.11902808e+00],
           [ 6.24863439e+00],
           ...,
           [ 2.13794980e-03],
           [-3.18195201e-01],
           [ 2.13794980e-03]])
    <font size="+1">Practice</font>  

    Practice

    Standardize the price column in "mexico-city-real-estate-1.csv":

    [9]:
    df1 = pd.read_csv("./data/mexico-city-real-estate-1.csv")
    [9]:
    operation property_type place_with_parent_names lat-lon price currency price_aprox_local_currency price_aprox_usd surface_total_in_m2 surface_covered_in_m2 price_usd_per_m2 price_per_m2 floor rooms expenses properati_url
    0 sell apartment |México|Distrito Federal|Álvaro Obregón| NaN 35000000.0 MXN 35634500.02 1894595.53 NaN NaN NaN NaN NaN NaN NaN http://alvaro-obregon.properati.com.mx/2eb_ven...
    1 sell apartment |México|Distrito Federal|Benito Juárez| NaN 2000000.0 MXN 2036257.11 108262.60 NaN NaN NaN NaN NaN NaN NaN http://benito-juarez.properati.com.mx/2ec_vent...
    2 sell apartment |México|Distrito Federal|Cuauhtémoc| 19.41501,-99.175174 2700000.0 MXN 2748947.10 146154.51 61.0 61.0 2395.975574 44262.295082 NaN 3.0 NaN http://cuauhtemoc.properati.com.mx/2pu_venta_a...
    3 sell apartment |México|Distrito Federal|Cuauhtémoc| 19.41501,-99.175174 6347000.0 MXN 6462061.92 343571.36 176.0 128.0 1952.110000 49585.937500 NaN 5.0 NaN http://cuauhtemoc.properati.com.mx/2pv_venta_a...
    4 sell apartment |México|Distrito Federal|Álvaro Obregón| NaN 6870000.0 MXN 6994543.16 371882.03 180.0 136.0 2066.011278 50514.705882 NaN 5.0 NaN http://alvaro-obregon.properati.com.mx/2pw_ven...
    [10]:
    X_transformed = ...
    [10]:
    Ellipsis
    ## One-Hot Encoding

    One-Hot Encoding¶

    A property's district is **categorical data**, or data which can be divided into groups.  For many machine learning algorithms, it's common to create a column in a DataFrame to indicate if the feature is present or absent, instead of using the category's name. First you a column for each district names then, for each observation, you put a 1 or a 0 to indicate if the property is located in each neighborhood or not. Let's take a look at the `mexico-city-test-features.csv` dataset for properties which include the district.

    A property's district is categorical data, or data which can be divided into groups. For many machine learning algorithms, it's common to create a column in a DataFrame to indicate if the feature is present or absent, instead of using the category's name. First you a column for each district names then, for each observation, you put a 1 or a 0 to indicate if the property is located in each neighborhood or not. Let's take a look at the mexico-city-test-features.csv dataset for properties which include the district.

    [11]:
    df = pd.read_csv("./data/mexico-city-test-features.csv").dropna()
    [11]:
    surface_covered_in_m2 lat lon neighborhood
    0 90.0 19.367931 -99.170262 Benito Juárez
    1 50.0 19.363542 -99.224084 Álvaro Obregón
    2 280.0 19.457982 -99.192690 Miguel Hidalgo
    3 55.0 19.334270 -99.083374 Iztapalapa
    4 80.0 19.416881 -99.109781 Venustiano Carranza
    You can do one-hot encoding using pandas [`get_dummies`](https://pandas.pydata.org/docs/reference/api/pandas.get_dummies.html) function, but we'll use a the [Category Encoders](https://contrib.scikit-learn.org/category_encoders/) library since it allows us to integrate the one hot encoder as a transformer in a scikit-learn Pipeline.

    You can do one-hot encoding using pandas get_dummies function, but we'll use a the Category Encoders library since it allows us to integrate the one hot encoder as a transformer in a scikit-learn Pipeline.

    [12]:
    from category_encoders import OneHotEncoder
    [12]:
    surface_covered_in_m2 lat lon neighborhood_Benito Juárez neighborhood_Álvaro Obregón neighborhood_Miguel Hidalgo neighborhood_Iztapalapa neighborhood_Venustiano Carranza neighborhood_Tlalpan neighborhood_Coyoacán neighborhood_La Magdalena Contreras neighborhood_Azcapotzalco neighborhood_Cuauhtémoc neighborhood_Cuajimalpa de Morelos neighborhood_Gustavo A. Madero neighborhood_Tláhuac neighborhood_Iztacalco neighborhood_Xochimilco
    0 90.0 19.367931 -99.170262 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
    1 50.0 19.363542 -99.224084 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
    2 280.0 19.457982 -99.192690 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0
    3 55.0 19.334270 -99.083374 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0
    4 80.0 19.416881 -99.109781 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0
    <font size="+1">Practice</font>  

    Practice

    Create a DataFrame which one-hot encodes the property_type column in mexico-city-real-estate-1.csv. The DataFrame you create should have extra columns for apartments, houses, and stores.

    [13]:
        "./data/mexico-city-real-estate-1.csv", usecols=["property_type"]
    ---------------------------------------------------------------------------
    AttributeError                            Traceback (most recent call last)
    Cell In [13], line 6
          4 ohe = ...
          5 mexico_city1_ohe = ...
    ----> 6 mexico_city1_ohe.head()
    
    AttributeError: 'ellipsis' object has no attribute 'head'
    ## Ordinal Encoding

    Ordinal Encoding¶

    For many machine learning algorithms, it's common to use one-hot encoding. This works well if there are a few categories, but as the number of features grows, the number of additional columns also grows. 

    For many machine learning algorithms, it's common to use one-hot encoding. This works well if there are a few categories, but as the number of features grows, the number of additional columns also grows.

    Having a large number of columns (and consequently a large number of features in your model) can lead to a number of issues often referred to as the curse of dimensionality. Two primary issues that can arise are computational complexity (operations performed on larger datasets may take longer) and overfitting (the model may not generalize to new data). In these scenarios, ordinal encoding is a popular choice for encoding the categorical variable. Instead of creating new columns, ordinal encoding simply replaces the categories in a categorical variable with integers.

    One potential risk of ordinal encoding is that some machine learning algorithms assume the integer values imply an ordering in the variables. This is important in logistic regression, where a relationship is defined between increases or decreases in the features and the target. Techniques like decision trees are okay to use ordinal encoding, because they generate splits. Rather than assuming any ordering between the numeric values, the splits will occur between the numeric values and effectively separate them. You can use the OrdinalEncoder transformer to perform ordinal encoding:

    [ ]:
    from category_encoders import OrdinalEncoder
    <font size="+1">Practice</font>  

    Practice

    Create a DataFrame which ordinal encodes the property_type column in mexico-city-real-estate-1.csv. The DataFrame you create should have integers replacing the values for apartments, houses, and stores.

    [ ]:
        "./data/mexico-city-real-estate-1.csv", usecols=["property_type"]
    ## Imputation

    Imputation¶

    Let's take a look at `mexico-city-real-estate-1.csv` and impute some of the missing values. First, we'll load the dataset, limiting ourselves to the `"surface_covered_in_m2"` and `"price_aprox_usd"` columns.

    Let's take a look at mexico-city-real-estate-1.csv and impute some of the missing values. First, we'll load the dataset, limiting ourselves to the "surface_covered_in_m2" and "price_aprox_usd" columns.

    [ ]:
    mexico_city1 = pd.read_csv("./data/mexico-city-real-estate-1.csv", usecols=columns)
    When you need to build a model using features that contain missing values, one helpful tool is the scikit-learn transformer [`SimpleImputer`](https://scikit-learn.org/stable/modules/generated/sklearn.impute.SimpleImputer.html). In order to use it, we need to start by instantiating the transformer. 

    When you need to build a model using features that contain missing values, one helpful tool is the scikit-learn transformer SimpleImputer. In order to use it, we need to start by instantiating the transformer.

    [ ]:
    from sklearn.impute import SimpleImputer
    Next, we train the imputer using the data. At this step it will calculate the mean value for each column.

    Next, we train the imputer using the data. At this step it will calculate the mean value for each column.

    [ ]:
    imputer.fit(mexico_city1)
    Last, we transform the data using the imputer.

    Last, we transform the data using the imputer.

    [ ]:
    mexico_city1_imputed = imputer.transform(mexico_city1)
    Since the imputer doesn't return a DataFrame, let's transform it into one. 

    Since the imputer doesn't return a DataFrame, let's transform it into one.

    [ ]:
    mexico_city1_imputed = pd.DataFrame(mexico_city1_imputed, columns=columns)
    Now there are no missing values!

    Now there are no missing values!

    Then we use the imputer to transform the data.

    Then we use the imputer to transform the data.

    Practice

    Read mexico-city-real-estate-1.csv into a DataFrame and impute the missing values for "surface_covered_in_m2" and "price_aprox_usd".WQU WorldQuant University Applied Data Science Lab QQQQ

    [ ]:
    mexico_city2_imputed = pd.DataFrame(mexico_city2_imputed, columns=columns)
    ## Data Leakage

    Data Leakage¶

    Let's consider the `mexico-city-real-estate-1.csv` dataset and fit a regression model using `surface_covered_in_m2` and `price_aprox_local_currency` to estimate `price_aprox_usd`.

    Let's consider the mexico-city-real-estate-1.csv dataset and fit a regression model using surface_covered_in_m2 and price_aprox_local_currency to estimate price_aprox_usd.

    [ ]:
    mexico_city1 = pd.read_csv("./data/mexico-city-real-estate-1.csv", usecols=columns)
    Now let's calculate the mean absolute error in our training data.

    Now let's calculate the mean absolute error in our training data.

    [ ]:
        mexico_city1[["surface_covered_in_m2", "price_aprox_local_currency"]]
    When you see a mean absolute error that's so close to zero (especially when the mean apartment price is so much larger), chances are there is leakage in your model!

    When you see a mean absolute error that's so close to zero (especially when the mean apartment price is so much larger), chances are there is leakage in your model!

    # Imbalanced Data

    Imbalanced Data¶

    When dealing with classification problems, we would ideally expect the training data to be evenly spread across different classes for better model performance. When the numbers of observations are uneven in different classes, we have imbalanced data. The class that represents the majority of observations is called the **majority class**, while the class with limited observation is called the **minority class**. Imbalanced data limits training data available for certain classes. In addition, when the one class takes the majority of the data, the model will keep predicting the majority class to achieve high accuracy result. Thus, prior to training a  model, it is essential to balance the data either through under-sampling the majority classes, or over-sampling the minority classes, or use other evaluation metrics like **recall** or **precision**.

    When dealing with classification problems, we would ideally expect the training data to be evenly spread across different classes for better model performance. When the numbers of observations are uneven in different classes, we have imbalanced data. The class that represents the majority of observations is called the majority class, while the class with limited observation is called the minority class. Imbalanced data limits training data available for certain classes. In addition, when the one class takes the majority of the data, the model will keep predicting the majority class to achieve high accuracy result. Thus, prior to training a model, it is essential to balance the data either through under-sampling the majority classes, or over-sampling the minority classes, or use other evaluation metrics like recall or precision.

    ## Under-sampling

    Under-sampling¶

    When data is imbalanced in different classes, one way we can balance it is reducing the number of observations in the majority class. This is called **under-sampling**. We can under-sample by randomly deleting some observations in the majority class. The open source [imbalanced-learn](https://imbalanced-learn.org/stable/) (imported as `imblearn`) is an open-source library that works with `scikit-learn` and provides tools when dealing with imbalanced classes. Here's an example of randomly deleting observations from the majority class using Poland bankruptcy data from 2008.

    When data is imbalanced in different classes, one way we can balance it is reducing the number of observations in the majority class. This is called under-sampling. We can under-sample by randomly deleting some observations in the majority class. The open source imbalanced-learn (imported as imblearn) is an open-source library that works with scikit-learn and provides tools when dealing with imbalanced classes. Here's an example of randomly deleting observations from the majority class using Poland bankruptcy data from 2008.

    [ ]:
    with gzip.open("data/poland-bankruptcy-data-2008.json.gz", "r") as f:
    The data is clearly imbalanced as there are many more observations in non-bankruptcy compared to bankruptcy.

    The data is clearly imbalanced as there are many more observations in non-bankruptcy compared to bankruptcy.

    [ ]:
    X, y = RandomUnderSampler().fit_resample(df[["company_id"]], df[["bankrupt"]])
    Now we have reduced the non-bankruptcy class to the same size as the bankruptcy class.

    Now we have reduced the non-bankruptcy class to the same size as the bankruptcy class.

    ## Over-sampling

    Over-sampling¶

    **Over-sampling** is the opposite of under-sampling. Instead of reducing the majority class, over-sampling increases the number of observations in the minority class by randomly making copies of the existing observations. Here is an example of making random copies from the minority class using the Poland bankruptcy data and `imblearn`.

    Over-sampling is the opposite of under-sampling. Instead of reducing the majority class, over-sampling increases the number of observations in the minority class by randomly making copies of the existing observations. Here is an example of making random copies from the minority class using the Poland bankruptcy data and imblearn.

    [ ]:
    X, y = RandomOverSampler().fit_resample(df[["company_id"]], df[["bankrupt"]])
    Now we have increased the bankruptcy class to the size of the non-bankruptcy class.

    Now we have increased the bankruptcy class to the size of the non-bankruptcy class.

    ### Practice

    Practice¶

    Now that you've seen an example of imbalanced data and how to under-  or over-sample it prior to model training, let's get some practice with the Poland bankruptcy data from 2007.

    Now that you've seen an example of imbalanced data and how to under- or over-sample it prior to model training, let's get some practice with the Poland bankruptcy data from 2007.

    [ ]:
    with gzip.open("data/poland-bankruptcy-data-2007.json.gz", "r") as f:
    First, check whether this data is imbalanced.

    First, check whether this data is imbalanced.

    [ ]:
    ​x
     
    Next, do under-sampling.

    Next, do under-sampling.

    [ ]:
    X, y = ...
    Finally, check whether the data is balanced.

    Finally, check whether the data is balanced.

    [ ]:
    ​x
     
    Great work! Now try over-sampling.

    Great work! Now try over-sampling.

    [ ]:
    X, y = ...
    And check whether the data is balanced.

    And check whether the data is balanced.

    [ ]:
    ​x
     
    # scikit-learn in Production

    scikit-learn in Production¶

    The previous examples have built models and made predictions one step at a time.  Many machine learning applications will require you to run the same steps many times, usually with new or updated data.  scikit-learn allows you to define a set of steps to process data for machine learning in a reproducible manner using a pipeline. 

    The previous examples have built models and made predictions one step at a time. Many machine learning applications will require you to run the same steps many times, usually with new or updated data. scikit-learn allows you to define a set of steps to process data for machine learning in a reproducible manner using a pipeline.

    ## Creating a Pipeline in scikit-learn

    Creating a Pipeline in scikit-learn¶

    First, we create a pipeline to do linear regression on the transformed data set.

    First, we create a pipeline to do linear regression on the transformed data set.

    [ ]:
    lin_reg = linear_model.LinearRegression()
    We can check the steps in the pipeline, but right now, there's only 1.

    We can check the steps in the pipeline, but right now, there's only 1.

    [ ]:
    pipe.named_steps
    Then we fit a linear regression model to our data.

    Then we fit a linear regression model to our data.

    [ ]:
    mexico_city1["surface_covered_in_m2"] = mexico_city1["surface_covered_in_m2"].astype(
    [ ]:
    print(y_pred.head())
    <font size="+1">Practice</font> 

    Practice

    Try this on the price_aprox_usd column in the mexico-city-real-estate-1.csv dataset.

    [ ]:
    print(y_pred.head())
    Let's use the `make_pipeline` function to create a pipeline to fit a linear regression model for the `mexico-city-real-estate-1.csv` dataset.

    Let's use the make_pipeline function to create a pipeline to fit a linear regression model for the mexico-city-real-estate-1.csv dataset.

    [ ]:
    X = mexico_city1.surface_covered_in_m2.values.reshape(-1, 1)
    Let's try to predict `price_aprox_usd` in the `mexico-city-test-features.csv` dataset.

    Let's try to predict price_aprox_usd in the mexico-city-test-features.csv dataset.

    [ ]:
    mexico_city_features = pd.read_csv("./data/mexico-city-test-features.csv")
    ## Accessing an Object in a Pipeline

    Accessing an Object in a Pipeline¶

    Let's figure out the regression coefficients.

    Let's figure out the regression coefficients.

    [ ]:
    pipe.named_steps["regressor"].coef_
    <font size="+1">Practice</font>

    Practice

    Now obtain the intercept

    [ ]:
    # INCLUDE pipe.named_steps[...].intercept_
    *References & Further Reading*

    References & Further Reading

    • One-Hot Encoding with the Category Encoder Package
    • Example of Using One-Hot Encoding
    • Online Example of Using One-Hot Encoding
    • Official pandas Documentation on Get Dummies
    • Online Tutorial on Pipelines for Linear Regression
    • scikit-learn Pipeline Documentation
    • Wikipedia article on the curse of dimensionality
    • Wikipedia Article on Leakage in Machine Learning
    • Official Pandas Documentation on Missing Data
    • Wikipedia Article on Imputation
    • Online Tutorial on Removing Rows with Missing Data
    • scikit-learn Documentation on SimpleImputer
    • imbalanced-learn Documentation
    ---

    Copyright 2022 WorldQuant University. This content is licensed solely for personal use. Redistribution or publication of this material is strictly prohibited.

    xxxxxxxxxx
    ​

    Usage Guidelines

    This lesson is part of the DS Lab core curriculum. For that reason, this notebook can only be used on your WQU virtual machine.

    This means:

    • ⓧ No downloading this notebook.
    • ⓧ No re-sharing of this notebook with friends or colleagues.
    • ⓧ No downloading the embedded videos in this notebook.
    • ⓧ No re-sharing embedded videos with friends or colleagues.
    • ⓧ No adding this notebook to public or private repositories.
    • ⓧ No uploading this notebook (or screenshots of it) to other websites, including websites for study resources.

    xxxxxxxxxx
    <font size="+3"><strong>Visualizing Data: plotly express</strong></font>

    Visualizing Data: plotly express

    xxxxxxxxxx
    There are many ways to interact with data, and one of the most powerful modes of interaction is through **visualizations**. Visualizations show data graphically, and are useful for exploring, analyzing, and presenting datasets. We use four libraries for making visualizations: [pandas](../%40textbook/07-visualization-pandas.ipynb), [Matplotlib](../%40textbook/06-visualization-matplotlib.ipynb), plotly express, and [seaborn](../%40textbook/09-visualization-seaborn.ipynb). In this section, we'll focus on using plotly express.

    There are many ways to interact with data, and one of the most powerful modes of interaction is through visualizations. Visualizations show data graphically, and are useful for exploring, analyzing, and presenting datasets. We use four libraries for making visualizations: pandas, Matplotlib, plotly express, and seaborn. In this section, we'll focus on using plotly express.

    xxxxxxxxxx
    # Scatter Plots

    Scatter Plots¶

    xxxxxxxxxx
    A **scatter plot** is a graph that uses dots to represent values for two different numeric variables. The position of each dot on the horizontal and vertical axis indicates values for an individual data point. Scatter plots are used to observe relationships between variables, and are especially useful if you're looking for **correlations**.

    A scatter plot is a graph that uses dots to represent values for two different numeric variables. The position of each dot on the horizontal and vertical axis indicates values for an individual data point. Scatter plots are used to observe relationships between variables, and are especially useful if you're looking for correlations.

    [1]:
    mexico_city1 = pd.read_csv("./data/mexico-city-real-estate-1.csv")
    [1]:
    operation property_type place_with_parent_names lat-lon price currency price_aprox_local_currency price_aprox_usd surface_total_in_m2 surface_covered_in_m2 price_per_m2 properati_url
    2 sell apartment |México|Distrito Federal|Cuauhtémoc| 19.41501,-99.175174 2700000.0 MXN 2748947.10 146154.51 61.0 61.0 44262.295082 http://cuauhtemoc.properati.com.mx/2pu_venta_a...
    3 sell apartment |México|Distrito Federal|Cuauhtémoc| 19.41501,-99.175174 6347000.0 MXN 6462061.92 343571.36 176.0 128.0 49585.937500 http://cuauhtemoc.properati.com.mx/2pv_venta_a...
    6 sell apartment |México|Distrito Federal|Miguel Hidalgo| 19.456564,-99.191724 670000.0 MXN 682146.11 36267.97 65.0 65.0 10307.692308 http://miguel-hidalgo-df.properati.com.mx/46h_...
    7 sell apartment |México|Distrito Federal|Gustavo A. Madero| 19.512787,-99.141393 1400000.0 MXN 1425379.97 75783.82 82.0 70.0 20000.000000 http://gustavo-a-madero.properati.com.mx/46p_v...
    8 sell house |México|Distrito Federal|Álvaro Obregón| 19.358776,-99.213557 6680000.0 MXN 6801098.67 361597.08 346.0 346.0 19306.358382 http://alvaro-obregon.properati.com.mx/46t_ven...
    xxxxxxxxxx
    After cleaning the data, we can use plotly express to draw scatter plots by specifying the DataFrame and the interested column names.

    After cleaning the data, we can use plotly express to draw scatter plots by specifying the DataFrame and the interested column names.

    [2]:
    fig = px.scatter(mexico_city1, x="price", y="surface_covered_in_m2")
    xxxxxxxxxx
    <font size="+1">Practice</font> 

    Practice

    Plot the scatter plot for column "price" and "surface_total_in_m2".

    [3]:
    fig = px.scatter(mexico_city1,x="price",y="surface_covered_in_m2")
    xxxxxxxxxx
    # 3D Scatter Plots

    3D Scatter Plots¶

    Scatter plots can summarize information in a DataFrame. Three dimensional scatter plots look great, but be careful: it can be difficult for people who might not be sure what they're looking at to accurately determine values of points in the plot. Still, scatter plots are useful for displaying relationships between three quantities that would be more difficult to observe in a two dimensional plot.

    Let's take a look at the first 50 rows of the mexico-city-real-estate-1.csv dataset.

    [4]:
    ] = mexico_city1["place_with_parent_names"].str.split("|", 4, expand=True)
    /tmp/ipykernel_659/3122900234.py:11: FutureWarning:
    
    In a future version of pandas all arguments of StringMethods.split except for the argument 'pat' will be keyword-only.
    
    
    xxxxxxxxxx
    Notice that the plot is interactive: you can rotate it zoom in or out. These kinds of plots also makes outliers easier to find; here, we can see that houses have higher prices than other types of properties.

    Notice that the plot is interactive: you can rotate it zoom in or out. These kinds of plots also makes outliers easier to find; here, we can see that houses have higher prices than other types of properties.

    xxxxxxxxxx
    <font size="+1">Practice</font> 

    Practice

    Modify the DataFrame to include columns for the base 10 log of price and surface_covered_in_m2 and then plot these for the entire mexico-city-real-estate-1.csv dataset.

    [5]:
    import math
    xxxxxxxxxx
    # Mapbox Scatter Plots

    Mapbox Scatter Plots¶

    xxxxxxxxxx
    A **mapbox scatter plot** is a special kind of scatter plot that allows you to create scatter plots in two dimensions and then superimpose them on top of a map. Our `mexico-city-real-estate-1.csv` dataset is a good place to start, because it includes **location data**. After importing the dataset and removing rows with missing data, split the `lat-lon` column into two separate columns: one for `latitude` and the other for `longitude`. Then use these to make a mapbox plot. Unfortunately, at present this type of plot does not easily allow for marker shape to vary based on a column of the DataFrame.

    A mapbox scatter plot is a special kind of scatter plot that allows you to create scatter plots in two dimensions and then superimpose them on top of a map. Our mexico-city-real-estate-1.csv dataset is a good place to start, because it includes location data. After importing the dataset and removing rows with missing data, split the lat-lon column into two separate columns: one for latitude and the other for longitude. Then use these to make a mapbox plot. Unfortunately, at present this type of plot does not easily allow for marker shape to vary based on a column of the DataFrame.

    [6]:
    mexico_city1[["latitude", "longitude"]] = mexico_city1["lat-lon"].str.split(
    /tmp/ipykernel_659/3692783844.py:6: FutureWarning:
    
    In a future version of pandas all arguments of StringMethods.split except for the argument 'pat' will be keyword-only.
    
    
    xxxxxxxxxx
    <font size="+1">Practice</font> 

    Practice

    Create another column in the DataFrame with a log scale of the prices. Then create three separate plots, one for stores, another for houses, and a final one for apartments. Color the points in the plots by the log of the price.

    [7]:
    from math import log10
    xxxxxxxxxx
    # Choropleth Maps

    Choropleth Maps¶

    xxxxxxxxxx
    A Choropleth Map is a map composed of colored polygons, showing the variable of interest at different color depth across geographies.Plotly express has a function called `px.choropleth` that be used to plot Choropleth Map. The challenges here are getting the geometry information. There are two ways, one is to use the built-in geometries in plotly when plot US States (use the state name directly) and world countries (use ISP-3 code). Another way is to look for GeoJSON files where each location has geometry information. In the following example, we will show the plot in US States with a synthetic data set.  

    A Choropleth Map is a map composed of colored polygons, showing the variable of interest at different color depth across geographies.Plotly express has a function called px.choropleth that be used to plot Choropleth Map. The challenges here are getting the geometry information. There are two ways, one is to use the built-in geometries in plotly when plot US States (use the state name directly) and world countries (use ISP-3 code). Another way is to look for GeoJSON files where each location has geometry information. In the following example, we will show the plot in US States with a synthetic data set.

    [8]:
        {"State": ["CA", "TX", "NY", "HI", "DE"], "Temparature": [100, 120, 110, 90, 105]}
    [8]:
    State Temparature
    0 CA 100
    1 TX 120
    2 NY 110
    3 HI 90
    4 DE 105
    [9]:
        df, locations="State", locationmode="USA-states", color="Temparature", scope="usa"
    xxxxxxxxxx
    # Histogram

    Histogram¶

    xxxxxxxxxx
    A **histogram** is a graph that shows the frequency distribution of numerical data. In addition to helping us understand frequency, histograms are also useful for detecting outliers. We can use the `px.histogram()` function from Plotly to draw histograms for specific columns, as long as the data type is numerical. Let's check the following example:

    A histogram is a graph that shows the frequency distribution of numerical data. In addition to helping us understand frequency, histograms are also useful for detecting outliers. We can use the px.histogram() function from Plotly to draw histograms for specific columns, as long as the data type is numerical. Let's check the following example:

    [10]:
    df = pd.read_csv("data/mexico-city-real-estate-1.csv")
    xxxxxxxxxx
    <font size="+1">Practice</font> 

    Practice

    Check the "surface_covered_in_m2" Histogram.

    [11]:
    fig = px.histogram(df,x="surface_covered_in_m2")
    xxxxxxxxxx
    # Boxplots

    Boxplots¶

    xxxxxxxxxx
    A **boxplot** is a graph that shows the minimum, first quartile, median, third quartile, and the maximum values in a dataset. Boxplots are useful because they provide a visual summary of the data, enabling researchers to quickly identify mean values, the dispersion of the data set, and signs of skewness. In the following example, we will explore how to draw boxplots for specific columns of a DataFrame.

    A boxplot is a graph that shows the minimum, first quartile, median, third quartile, and the maximum values in a dataset. Boxplots are useful because they provide a visual summary of the data, enabling researchers to quickly identify mean values, the dispersion of the data set, and signs of skewness. In the following example, we will explore how to draw boxplots for specific columns of a DataFrame.

    [12]:
    mexico_city1 = pd.read_csv("./data/mexico-city-real-estate-1.csv")
    [12]:
    operation property_type place_with_parent_names lat-lon price currency price_aprox_local_currency price_aprox_usd surface_total_in_m2 surface_covered_in_m2 price_per_m2 properati_url
    2 sell apartment |México|Distrito Federal|Cuauhtémoc| 19.41501,-99.175174 2700000.0 MXN 2748947.10 146154.51 61.0 61.0 44262.295082 http://cuauhtemoc.properati.com.mx/2pu_venta_a...
    3 sell apartment |México|Distrito Federal|Cuauhtémoc| 19.41501,-99.175174 6347000.0 MXN 6462061.92 343571.36 176.0 128.0 49585.937500 http://cuauhtemoc.properati.com.mx/2pv_venta_a...
    6 sell apartment |México|Distrito Federal|Miguel Hidalgo| 19.456564,-99.191724 670000.0 MXN 682146.11 36267.97 65.0 65.0 10307.692308 http://miguel-hidalgo-df.properati.com.mx/46h_...
    7 sell apartment |México|Distrito Federal|Gustavo A. Madero| 19.512787,-99.141393 1400000.0 MXN 1425379.97 75783.82 82.0 70.0 20000.000000 http://gustavo-a-madero.properati.com.mx/46p_v...
    8 sell house |México|Distrito Federal|Álvaro Obregón| 19.358776,-99.213557 6680000.0 MXN 6801098.67 361597.08 346.0 346.0 19306.358382 http://alvaro-obregon.properati.com.mx/46t_ven...
    xxxxxxxxxx
    Check the boxplot for column `"price"`:

    Check the boxplot for column "price":

    [13]:
    fig = px.box(mexico_city1, y="price")
    xxxxxxxxxx
    If you want to check the distribution of a column value by different categories, defined by another categorical column, you can add an `x` argument to specify the name of the categorical column. In the following example, we check the price distribution across different property types:

    If you want to check the distribution of a column value by different categories, defined by another categorical column, you can add an x argument to specify the name of the categorical column. In the following example, we check the price distribution across different property types:

    [14]:
    fig = px.box(mexico_city1, x="property_type", y="price")
    xxxxxxxxxx
    <font size="+1">Practice</font> 

    Practice

    Check the "surface_covered_in_m2" distribution by property types.

    [15]:
    fig.show()
    ---------------------------------------------------------------------------
    AttributeError                            Traceback (most recent call last)
    Cell In [15], line 2
          1 fig = ...
    ----> 2 fig.show()
    
    AttributeError: 'ellipsis' object has no attribute 'show'
    xxxxxxxxxx
    # Bar Chart

    Bar Chart¶

    xxxxxxxxxx
    A **bar chart** is a graph that shows all the values of a categorical variable in a dataset. They consist of an axis and a series of labeled horizontal or vertical bars. The bars depict frequencies of different values of a variable or simply the different values themselves. The numbers on the y-axis of a vertical bar chart or the x-axis of a horizontal bar chart are called the scale. 

    A bar chart is a graph that shows all the values of a categorical variable in a dataset. They consist of an axis and a series of labeled horizontal or vertical bars. The bars depict frequencies of different values of a variable or simply the different values themselves. The numbers on the y-axis of a vertical bar chart or the x-axis of a horizontal bar chart are called the scale.

    In the following example, we will see some bar plots based on the Mexico City real estate dataset. Specifically, we will count the number of observations in each borough and plot them. We first need to read the data set and extract Borough and other location information from column "place_with_parent_names".

    [ ]:
    ] = mexico_city1["place_with_parent_names"].str.split("|", 4, expand=True)
    xxxxxxxxxx
    We can calculate the number of real estate showing in the data set by Borough using `value_counts()`, then plot it as bar plot:

    We can calculate the number of real estate showing in the data set by Borough using value_counts(), then plot it as bar plot:

    [ ]:
    mexico_city1["Borough"].value_counts()
    [ ]:
    fig = px.bar(mexico_city1["Borough"].value_counts())
    xxxxxxxxxx
    We can plot more expressive bar plots by adding more arguments. For example, we can plot the number of observations by borough and property type. First of all, we need use `groupby` to calculate the aggregated counts for each Borough and property type combination:

    We can plot more expressive bar plots by adding more arguments. For example, we can plot the number of observations by borough and property type. First of all, we need use groupby to calculate the aggregated counts for each Borough and property type combination:

    [ ]:
    size_df = mexico_city1.groupby(["Borough", "property_type"], as_index=False).size()
    xxxxxxxxxx
    By specifying `x`, `y` and `color`, the following bar graph shows the total counts by Borough, with different property types showing in different colors. Note `y` has to be numerical, while `x` and `color` are usually categorical variables.<span style='color: transparent; font-size:1%'>WQU WorldQuant University Applied Data Science Lab QQQQ</span>

    By specifying x, y and color, the following bar graph shows the total counts by Borough, with different property types showing in different colors. Note y has to be numerical, while x and color are usually categorical variables.WQU WorldQuant University Applied Data Science Lab QQQQ

    [ ]:
    fig = px.bar(size_df, x="Borough", y="size", color="property_type", barmode="relative")
    xxxxxxxxxx
    Note the argument `barmode` is specified as 'relative', which is also the default value. In this mode, bars are stacked above each other. We can also use 'overlay' where bars are drawn on top of each other.

    Note the argument barmode is specified as 'relative', which is also the default value. In this mode, bars are stacked above each other. We can also use 'overlay' where bars are drawn on top of each other.

    [ ]:
    fig = px.bar(size_df, x="Borough", y="size", color="property_type", barmode="overlay")
    xxxxxxxxxx
    If we want bars to be placed beside each other, we can specify `barmode` as "group":

    If we want bars to be placed beside each other, we can specify barmode as "group":

    [ ]:
    fig = px.bar(size_df, x="Borough", y="size", color="property_type", barmode="group")
    xxxxxxxxxx
    <font size="+1">Practice</font> 

    Practice

    Plot bar plot for the number of observations by property types in "mexico-city-real-estate-1.csv".

    [ ]:
    bar_df = ...
    xxxxxxxxxx
    # References and Further Reading

    References and Further Reading¶

    • Official plotly express Documentation on Scatter Plots
    • Official plotly Express Documentation on 3D Plots
    • Official plotly Documentation on Notebooks
    • plotly Community Forum Post on Axis Labeling
    • plotly express Official Documentation on Tile Maps
    • plotly Choropleth Maps in Python Document
    • plotly express Official Documentation on Figure Display
    • Online Tutorial on String Conversion in Pandas
    • Official Pandas Documentation on using Lambda Functions on a Column
    • Official Seaborn Documentation on Generating a Heatmap
    • Online Tutorial on Correlation Matrices in Pandas
    • Official Pandas Documentation on Correlation Matrices
    • Official Matplotlib Documentation on Colormaps
    • Official Pandas Documentation on Box Plots
    • Online Tutorial on Box Plots
    • Online Tutorial on Axes Labels in Seaborn and Matplotlib
    • Matplotlib Gallery Example of an Annotated Heatmap
    xxxxxxxxxx
    ---

    Copyright 2022 WorldQuant University. This content is licensed solely for personal use. Redistribution or publication of this material is strictly prohibited.

    ​x
     

    Usage Guidelines

    This lesson is part of the DS Lab core curriculum. For that reason, this notebook can only be used on your WQU virtual machine.

    This means:

    • ⓧ No downloading this notebook.
    • ⓧ No re-sharing of this notebook with friends or colleagues.
    • ⓧ No downloading the embedded videos in this notebook.
    • ⓧ No re-sharing embedded videos with friends or colleagues.
    • ⓧ No adding this notebook to public or private repositories.
    • ⓧ No uploading this notebook (or screenshots of it) to other websites, including websites for study resources.

    <font size="+3"><strong>Databases: PyMongo</strong></font>

    Databases: PyMongo

    # Working with PyMongo

    Working with PyMongo¶

    For all of these examples, we're going to be working with the "lagos" collection in the "air-quality" database. Before we can do anything else, we need to bring in pandas (which we won't use until the very end), pprint (a module that lets us see the data in an understandable way), and PyMongo (a library for working with MongoDB databases).

    [1]:
    from pprint import PrettyPrinter
    ## Databases

    Databases¶

    Data comes to us in lots of different ways, and one of those ways is in a database. A database is a collection of data.

    ## Servers and Clients

    Servers and Clients¶

    Next, we need to connect to a server. A database server is where the database resides; it can be accessed using a client. Without a client, a database is just a collection of information that we can't work with, because we have no way in. We're going to be learning more about a database called MongoDB, and we'll use PrettyPrinter to make the information it generates easier to understand. Here's how the connection works:

    [2]:
    client = MongoClient(host="localhost", port=27017)
    ## Semi-structured Data

    Semi-structured Data¶

    Databases are designed to work with either structured data or semi-structured data. In this part of the course, we're going to be working with databases that contain semi-structured data. Data is semi-structured when it has some kind of organizing logic, but that logic can't be displayed using rows and columns. Your email account contains semi-structured data if it’s divided into sections like Inbox, Sent, and Trash. If you’ve ever seen tweets from Twitter grouped by hashtag, that’s semi-structured data too. Semi-structured data is also used in sensor readings, which is what we'll be working with here.

    ## Exploring a Database

    Exploring a Database¶

    So, now that we're connected to a server, let's take a look at what's there. Working our way down the specificity scale, the first thing we need to do is figure out which databases are on this server. To see which databases on the server, we'll use the list_databases method, like this:

    [3]:
    pp.pprint(list(client.list_databases()))
    [ {'empty': False, 'name': 'admin', 'sizeOnDisk': 40960},
      {'empty': False, 'name': 'air-quality', 'sizeOnDisk': 7000064},
      {'empty': False, 'name': 'config', 'sizeOnDisk': 12288},
      {'empty': False, 'name': 'local', 'sizeOnDisk': 73728},
      {'empty': False, 'name': 'wqu-abtest', 'sizeOnDisk': 585728}]
    
    It looks like this server contains four databases: `"admin"`, `"air-quality"`, `"config"`, and `"local"`. We're only interested in `"air-quality"`, so let's connect to that one:

    It looks like this server contains four databases: "admin", "air-quality", "config", and "local". We're only interested in "air-quality", so let's connect to that one:

    [4]:
    db = client["air-quality"]
    In MongoDB, a **database** is a container for **collections**. Each database gets its own set of files, and a single MongoDB **server** typically has multiple databases.

    In MongoDB, a database is a container for collections. Each database gets its own set of files, and a single MongoDB server typically has multiple databases.

    ## Collections

    Collections¶

    Let's use a for loop to take a look at the collections in the "air-quality" database:

    [5]:
    for c in db.list_collections():
    system.views
    lagos
    system.buckets.lagos
    nairobi
    system.buckets.nairobi
    dar-es-salaam
    system.buckets.dar-es-salaam
    
    As you can see, there are three actual collections here: `"nairobi"`, `"lagos"`, and `"dar-es-salaam"`. Since we're only interested in the `"lagos"` collection, let's get it on its own like this: 

    As you can see, there are three actual collections here: "nairobi", "lagos", and "dar-es-salaam". Since we're only interested in the "lagos" collection, let's get it on its own like this:

    [6]:
    lagos = db["lagos"]
    ## Documents

    Documents¶

    A MongoDB **document** is an individual record of data in a **collection**, and is the basic unit of analysis in MongoDB. Documents come with **metadata** that helps us understand what the document is; we'll get back to that in a minute. In the meantime, let's use the [`count_documents`](https://pymongo.readthedocs.io/en/stable/api/pymongo/collection.html#pymongo.collection.Collection.count_documents) method to see how many documents the `"lagos"` collection contains:

    A MongoDB document is an individual record of data in a collection, and is the basic unit of analysis in MongoDB. Documents come with metadata that helps us understand what the document is; we'll get back to that in a minute. In the meantime, let's use the count_documents method to see how many documents the "lagos" collection contains:

    [7]:
    lagos.count_documents({})
    [7]:
    166496
    <font size="+1">Practice</font>

    Practice

    Try it yourself! Bring in all the necessary libraries and modules, then connect to the "air-quality" database and print the number of documents in the "nairobi" collection.

    [8]:
    client = MongoClient(host = "localhost", port = 27017)
    [    {'empty': False, 'name': 'admin', 'sizeOnDisk': 40960},
         {'empty': False, 'name': 'air-quality', 'sizeOnDisk': 7000064},
         {'empty': False, 'name': 'config', 'sizeOnDisk': 12288},
         {'empty': False, 'name': 'local', 'sizeOnDisk': 73728},
         {'empty': False, 'name': 'wqu-abtest', 'sizeOnDisk': 585728}]
    system.views
    lagos
    system.buckets.lagos
    nairobi
    system.buckets.nairobi
    dar-es-salaam
    system.buckets.dar-es-salaam
    
    [8]:
    202212
    ### Retrieving Data

    Retrieving Data¶

    Now that we know how many documents the "lagos" collection contains, let's take a closer look at what's there. The first thing you'll notice is that the output starts out with a curly bracket ({), and ends with a curly bracket (}). That tells us that this information is a dictionary. To access documents in the collection, we'll use two methods: find and find_one. As you might expect, find will retrieve all the documents, and find_one will bring back only the first document. For now, let's stick to find_one; we'll come back to find later.

    Just like everywhere else, we'll need to assign a variable name to whatever comes back, so let's call this one result.

    [9]:
    result = lagos.find_one({})
    {    '_id': ObjectId('6334b0f18c51459f9b1d955d'),
         'metadata': {    'lat': 6.501,
                          'lon': 3.367,
                          'measurement': 'temperature',
                          'sensor_id': 10,
                          'sensor_type': 'DHT11',
                          'site': 4},
         'temperature': nan,
         'timestamp': datetime.datetime(2018, 1, 7, 7, 7, 3, 88000)}
    
    ### Key-Value Pairs

    Key-Value Pairs¶

    There's a lot going on here! Let's work from the bottom up, starting with this:

    {
        'temperature': 27.0,
        'timestamp': datetime.datetime(2017, 9, 6, 13, 18, 10, 120000)
    }
    

    The actual data is labeled temperature and timestamp, and if seeing it presented this way seems familiar, that's because what you're seeing at the bottom are two key-value pairs. In PyMongo, "_id" is always the primary key. Primary keys are the column(s) which contain values that uniquely identify each row in a table; we'll talk about that more in a minute.

    ### Metadata

    Metadata¶

    Next, we have this:

    'metadata': { 'lat': 6.602,
                  'lon': 3.351,
                  'measurement': 'temperature',
                  'sensor_id': 9,
                  'sensor_type': 'DHT11',
                  'site': 2}
    

    This is the document's metadata. Metadata is data about the data. If you’re working with a database, its data is the information it contains, and its metadata describes what that information is. In MongoDB, each document often has metadata of its own. If we go back to the example of your email account, each message in your Sent folder includes both the message itself and information about when you sent it and who you sent it to; the message is data, and the other information is metadata.

    The metadata we see in this block of code tells us what the key-value pairs from the last code block mean, and where the information stored there comes from. There's location data, a line telling us what about the format of the key-value pairs, some information about the equipment used to gather the data, and where the data came from.

    ### Identifiers

    Identifiers¶

    Finally, at the top, we have this:

    { 
        '_id': ObjectId('6126f1780e45360640bf240a')
    }
    

    This is the document's unique identifier, which is similar to the index label for each row in a pandas DataFrame.

    <font size="+1">Practice</font>

    Practice

    Try it yourself! Retrieve a single document from the "nairobi" collection, and print the result.

    [10]:
    result = nairobi.find_one({})
    {    'P1': 39.67,
         '_id': ObjectId('6334b0e98c51459f9b198d27'),
         'metadata': {    'lat': -1.3,
                          'lon': 36.785,
                          'measurement': 'P1',
                          'sensor_id': 57,
                          'sensor_type': 'SDS011',
                          'site': 29},
         'timestamp': datetime.datetime(2018, 9, 1, 0, 0, 2, 472000)}
    
    ## Analyzing Data

    Analyzing Data¶

    Now that we've seen what a document looks like in this collection, let's start working with what we've got. Since our metadata includes information about each sensor's "site", we might be curious to know how many sites are in the "lagos" collection. To do that, we'll use the distinct method, like this:

    [11]:
    lagos.distinct("metadata.site")
    [11]:
    [3, 4]
    Notice that in order to grab the `"site"` number, we needed to include the `"metadata"` tag. 

    Notice that in order to grab the "site" number, we needed to include the "metadata" tag.

    This tells us that there are 2 sensor sites in Lagos: one labeled 3 and the other labeled 4.

    Let's go further. We know that there are two sensor sites in Lagos, but we don't know how many documents are associated with each site. To find that out, we'll use the count_documents method for each site.

    [12]:
    print("Documents from site 3:", lagos.count_documents({"metadata.site": 3}))
    Documents from site 3: 140586
    Documents from site 4: 25910
    
    <font size="+1">Practice</font>

    Practice

    Try it yourself! Find out how many sensor sites are in Nairobi, what their labels are, and how many documents are associated with each one.

    [13]:
    print("Documents from site 29:", nairobi.count_documents({"metadata.site":29}))
    Documents from site 29: 131852
    Documents from site 6: 70360
    
    [14]:
    print("Documents from site 29:", nairobi.count_documents({"metadata.site": 29}))
    Documents from site 29: 131852
    Documents from site 6: 70360
    
    Now that we know how many *documents* are associated with each site, let's keep drilling down and find the number of *readings* for each site. We'll do this with the [`aggregate`](https://pymongo.readthedocs.io/en/stable/api/pymongo/collection.html#pymongo.collection.Collection.aggregate) method.

    Now that we know how many documents are associated with each site, let's keep drilling down and find the number of readings for each site. We'll do this with the aggregate method.

    Before we run it, let's take a look at some of what's happening in the code here. First, you'll notice that there are several dollar signs ($) in the list. This is telling the collection that we want to create something new. Here, we're saying that we want there to be a new group, and that the new group needs to be updated with data from metadata.site, and then updated again with data from count.

    There's also a new field: "_id". In PyMongo, "_id" is always the primary key. Primary keys are the fields which contain values that uniquely identify each row in a table.

    Let's run the code and see what happens:

    [15]:
        [{"$group": {"_id": "$metadata.site", "count": {"$count": {}}}}]
    [{'_id': 3, 'count': 140586}, {'_id': 4, 'count': 25910}]
    
    With that information in mind, we might want to know what those readings actually are. Since we're really interested in measures of air quality, let's take a look at the `P2` values in the `"lagos"` collection. `P2` measures the amount of particulate matter in the air, which in this case is something called PM 2.5. If we wanted to get all the documents in a collection, we could, but that would result in an unmanageably large number of records clogging up the memory on our machines. Instead, let's use the [`find`](https://pymongo.readthedocs.io/en/stable/api/pymongo/collection.html#pymongo.collection.Collection.find) method and use `limit` to make sure we only get back the first 3. 

    With that information in mind, we might want to know what those readings actually are. Since we're really interested in measures of air quality, let's take a look at the P2 values in the "lagos" collection. P2 measures the amount of particulate matter in the air, which in this case is something called PM 2.5. If we wanted to get all the documents in a collection, we could, but that would result in an unmanageably large number of records clogging up the memory on our machines. Instead, let's use the find method and use limit to make sure we only get back the first 3.

    [16]:
    result = lagos.find({"metadata.measurement": "P2"}).limit(3)
    [    {    'P2': 14.42,
              '_id': ObjectId('6334b0f28c51459f9b1de145'),
              'metadata': {    'lat': 6.501,
                               'lon': 3.367,
                               'measurement': 'P2',
                               'sensor_id': 6,
                               'sensor_type': 'PPD42NS',
                               'site': 4},
              'timestamp': datetime.datetime(2018, 1, 7, 7, 7, 3, 39000)},
         {    'P2': 19.66,
              '_id': ObjectId('6334b0f28c51459f9b1de146'),
              'metadata': {    'lat': 6.501,
                               'lon': 3.367,
                               'measurement': 'P2',
                               'sensor_id': 6,
                               'sensor_type': 'PPD42NS',
                               'site': 4},
              'timestamp': datetime.datetime(2018, 1, 7, 7, 11, 23, 870000)},
         {    'P2': 24.79,
              '_id': ObjectId('6334b0f28c51459f9b1de147'),
              'metadata': {    'lat': 6.501,
                               'lon': 3.367,
                               'measurement': 'P2',
                               'sensor_id': 6,
                               'sensor_type': 'PPD42NS',
                               'site': 4},
              'timestamp': datetime.datetime(2018, 1, 7, 7, 21, 53, 981000)}]
    
    <font size="+1">Practice</font>

    Practice

    Try it yourself! Find out how many sensor sites are in Nairobi, what their labels are, how many documents are associated with each one, and the number of observations from each site. Then, return the first three documents with the value P2.

    [17]:
        [{"$group": {"_id": "$metadata.site", "count": {"$count": {}}}}]
    [{'_id': 6, 'count': 70360}, {'_id': 29, 'count': 131852}]
    [    {    'P2': 14.42,
              '_id': ObjectId('6334b0f28c51459f9b1de145'),
              'metadata': {    'lat': 6.501,
                               'lon': 3.367,
                               'measurement': 'P2',
                               'sensor_id': 6,
                               'sensor_type': 'PPD42NS',
                               'site': 4},
              'timestamp': datetime.datetime(2018, 1, 7, 7, 7, 3, 39000)},
         {    'P2': 19.66,
              '_id': ObjectId('6334b0f28c51459f9b1de146'),
              'metadata': {    'lat': 6.501,
                               'lon': 3.367,
                               'measurement': 'P2',
                               'sensor_id': 6,
                               'sensor_type': 'PPD42NS',
                               'site': 4},
              'timestamp': datetime.datetime(2018, 1, 7, 7, 11, 23, 870000)},
         {    'P2': 24.79,
              '_id': ObjectId('6334b0f28c51459f9b1de147'),
              'metadata': {    'lat': 6.501,
                               'lon': 3.367,
                               'measurement': 'P2',
                               'sensor_id': 6,
                               'sensor_type': 'PPD42NS',
                               'site': 4},
              'timestamp': datetime.datetime(2018, 1, 7, 7, 21, 53, 981000)}]
    
    So far, we've been dealing with relatively small subsets of the data in our collections, but what if we need to work with something bigger? Let's start by using `distinct` to remind ourselves of the kinds of data we have at our disposal.

    So far, we've been dealing with relatively small subsets of the data in our collections, but what if we need to work with something bigger? Let's start by using distinct to remind ourselves of the kinds of data we have at our disposal.

    [18]:
    lagos.distinct("metadata.measurement")
    [18]:
    ['humidity', 'temperature', 'P1', 'P2']
    There are also comparison query operators that can be helpful for filtering the data. In total, we have 

    There are also comparison query operators that can be helpful for filtering the data. In total, we have

    • $gt: greater than (>)
    • $lt: less than (<)
    • $gte: greater than equal to (>=)
    • $lte: less than equal to (<= )

    Let's use the timestamp to see how we can use these operators to select different documents:

    [19]:
    result = nairobi.find({"timestamp": {"$gt": datetime.datetime(2018, 9, 1)}}).limit(3)
    [    {    'P1': 39.67,
              '_id': ObjectId('6334b0e98c51459f9b198d27'),
              'metadata': {    'lat': -1.3,
                               'lon': 36.785,
                               'measurement': 'P1',
                               'sensor_id': 57,
                               'sensor_type': 'SDS011',
                               'site': 29},
              'timestamp': datetime.datetime(2018, 9, 1, 0, 0, 2, 472000)},
         {    'P1': 39.13,
              '_id': ObjectId('6334b0e98c51459f9b198d28'),
              'metadata': {    'lat': -1.3,
                               'lon': 36.785,
                               'measurement': 'P1',
                               'sensor_id': 57,
                               'sensor_type': 'SDS011',
                               'site': 29},
              'timestamp': datetime.datetime(2018, 9, 1, 0, 5, 3, 941000)},
         {    'P1': 30.07,
              '_id': ObjectId('6334b0e98c51459f9b198d29'),
              'metadata': {    'lat': -1.3,
                               'lon': 36.785,
                               'measurement': 'P1',
                               'sensor_id': 57,
                               'sensor_type': 'SDS011',
                               'site': 29},
              'timestamp': datetime.datetime(2018, 9, 1, 0, 10, 4, 374000)}]
    
    [20]:
    result = nairobi.find({"timestamp": {"$lt": datetime.datetime(2018, 12, 1)}}).limit(3)
    [    {    'P1': 39.67,
              '_id': ObjectId('6334b0e98c51459f9b198d27'),
              'metadata': {    'lat': -1.3,
                               'lon': 36.785,
                               'measurement': 'P1',
                               'sensor_id': 57,
                               'sensor_type': 'SDS011',
                               'site': 29},
              'timestamp': datetime.datetime(2018, 9, 1, 0, 0, 2, 472000)},
         {    'P1': 39.13,
              '_id': ObjectId('6334b0e98c51459f9b198d28'),
              'metadata': {    'lat': -1.3,
                               'lon': 36.785,
                               'measurement': 'P1',
                               'sensor_id': 57,
                               'sensor_type': 'SDS011',
                               'site': 29},
              'timestamp': datetime.datetime(2018, 9, 1, 0, 5, 3, 941000)},
         {    'P1': 30.07,
              '_id': ObjectId('6334b0e98c51459f9b198d29'),
              'metadata': {    'lat': -1.3,
                               'lon': 36.785,
                               'measurement': 'P1',
                               'sensor_id': 57,
                               'sensor_type': 'SDS011',
                               'site': 29},
              'timestamp': datetime.datetime(2018, 9, 1, 0, 10, 4, 374000)}]
    
    [21]:
        {"timestamp": {"$eq": datetime.datetime(2018, 9, 1, 0, 0, 2, 472000)}}
    [    {    'P1': 39.67,
              '_id': ObjectId('6334b0e98c51459f9b198d27'),
              'metadata': {    'lat': -1.3,
                               'lon': 36.785,
                               'measurement': 'P1',
                               'sensor_id': 57,
                               'sensor_type': 'SDS011',
                               'site': 29},
              'timestamp': datetime.datetime(2018, 9, 1, 0, 0, 2, 472000)},
         {    'P2': 34.43,
              '_id': ObjectId('6334b0ea8c51459f9b1a0db2'),
              'metadata': {    'lat': -1.3,
                               'lon': 36.785,
                               'measurement': 'P2',
                               'sensor_id': 57,
                               'sensor_type': 'SDS011',
                               'site': 29},
              'timestamp': datetime.datetime(2018, 9, 1, 0, 0, 2, 472000)}]
    
    <font size="+1">Practice</font>

    Practice

    Try it yourself! Find three documents with timestamp greater than or equal to and less than or equal the date December 12, 2018 — datetime.datetime(2018, 12, 1, 0, 0, 6, 767000).

    [22]:
    result = nairobi.find({"timestamp":{"$gte":datetime.datetime(2018,12,1,0,0,6,767000)}}).limit(3)
    [    {    'P1': 17.08,
              '_id': ObjectId('6334b0e98c51459f9b19eba8'),
              'metadata': {    'lat': -1.3,
                               'lon': 36.785,
                               'measurement': 'P1',
                               'sensor_id': 57,
                               'sensor_type': 'SDS011',
                               'site': 29},
              'timestamp': datetime.datetime(2018, 12, 1, 0, 0, 6, 767000)},
         {    'P1': 17.62,
              '_id': ObjectId('6334b0e98c51459f9b19eba9'),
              'metadata': {    'lat': -1.3,
                               'lon': 36.785,
                               'measurement': 'P1',
                               'sensor_id': 57,
                               'sensor_type': 'SDS011',
                               'site': 29},
              'timestamp': datetime.datetime(2018, 12, 1, 0, 5, 6, 327000)},
         {    'P1': 11.05,
              '_id': ObjectId('6334b0e98c51459f9b19ebaa'),
              'metadata': {    'lat': -1.3,
                               'lon': 36.785,
                               'measurement': 'P1',
                               'sensor_id': 57,
                               'sensor_type': 'SDS011',
                               'site': 29},
              'timestamp': datetime.datetime(2018, 12, 1, 0, 10, 5, 579000)}]
    
    [23]:
    # Less than or equal to
    ---------------------------------------------------------------------------
    TypeError                                 Traceback (most recent call last)
    Cell In [23], line 5
          1 # Less than or equal to
          3 result = ...
    ----> 5 pp.pprint(list(result))
    
    TypeError: 'ellipsis' object is not iterable
    ## Updating Documents

    Updating Documents¶

    We can also update documents by passing some filter and new values using `update_one` to update one record or `update_many` to update many records. Let's look at an example. Before updating, we have this record showing like this:

    We can also update documents by passing some filter and new values using update_one to update one record or update_many to update many records. Let's look at an example. Before updating, we have this record showing like this:

    [ ]:
        {"timestamp": {"$eq": datetime.datetime(2018, 9, 1, 0, 0, 2, 472000)}}
    Now we are updating the sensor type from `"SDS011"` to `"SDS"`, we first select all records with sensor type equal to `"SDS011"`, then set the new value to `"SDS"`:

    Now we are updating the sensor type from "SDS011" to "SDS", we first select all records with sensor type equal to "SDS011", then set the new value to "SDS":

    [ ]:
        {"metadata.sensor_type": {"$eq": "SDS101"}},
    Now we can see all records have changed:

    Now we can see all records have changed:

    [ ]:
        {"timestamp": {"$eq": datetime.datetime(2018, 9, 1, 0, 0, 2, 472000)}}
    We can change it back:

    We can change it back:

    [ ]:
        {"$set": {"metadata.sensor_type": "SDS101"}},
    [ ]:
    result.raw_result
    ## Aggregation

    Aggregation¶

    Since we're looking for *big* numbers, we need to figure out which one of those dimensions has the largest number of measurements by **aggregating** the data in each document. Since we already know that `site 3` has significantly more documents than `site 2`, let's start looking at `site 3`. We can use the `$match` syntax to only select `site 3` data. The code to do that looks like this: 

    Since we're looking for big numbers, we need to figure out which one of those dimensions has the largest number of measurements by aggregating the data in each document. Since we already know that site 3 has significantly more documents than site 2, let's start looking at site 3. We can use the $match syntax to only select site 3 data. The code to do that looks like this:

    [ ]:
            {"$group": {"_id": "$metadata.measurement", "count": {"$count": {}}}},
    <font size="+1">Practice</font>

    Practice

    Try it yourself! Find the number of each measurement type at site 29 in Nairobi.

    [ ]:
    pp.pprint(list(result))
    After aggregation, there is another useful operator called `$project`, which allows you to specify which fields to display by adding new fields or deleting fields. Using the Nairobi data from site 29, we can first count each sensor type:<span style='color: transparent; font-size:1%'>WQU WorldQuant University Applied Data Science Lab QQQQ</span>

    After aggregation, there is another useful operator called $project, which allows you to specify which fields to display by adding new fields or deleting fields. Using the Nairobi data from site 29, we can first count each sensor type:WQU WorldQuant University Applied Data Science Lab QQQQ

    [ ]:
            {"$group": {"_id": "$metadata.sensor_type", "count": {"$count": {}}}},
    We can see there are two sensor types and the corresponding counts. If we only want to display what are the types but do not care about the counts, we can suppress the `count` filed by setting it at 0 in `$project`:

    We can see there are two sensor types and the corresponding counts. If we only want to display what are the types but do not care about the counts, we can suppress the count filed by setting it at 0 in $project:

    [ ]:
            {"$group": {"_id": "$metadata.sensor_type", "count": {"$count": {}}}},
    The `$project` syntax is also useful for deleting the intermediate fields that we used to generate our final fields but no longer need. In the following example, let's calculate the date difference for each sensor type. We'll first use the aggregation method to get the start date and last date. 

    The $project syntax is also useful for deleting the intermediate fields that we used to generate our final fields but no longer need. In the following example, let's calculate the date difference for each sensor type. We'll first use the aggregation method to get the start date and last date.

    [ ]:
                    "date_min": {"$min": "$timestamp"},
    Then we can calculate the date difference using `$dateDiff`, which gets the date difference through specifying the start date, end date and unit for timestamp data. We can see from the results above that the dates, are very close to each other. The only differences are in the minutes, so we can specify the unit as minute to show the difference. Since we don't need the start date and end dates, we can define a `"dateDiff"` field inside `$project`, so that it will be shown in the final display: 

    Then we can calculate the date difference using $dateDiff, which gets the date difference through specifying the start date, end date and unit for timestamp data. We can see from the results above that the dates, are very close to each other. The only differences are in the minutes, so we can specify the unit as minute to show the difference. Since we don't need the start date and end dates, we can define a "dateDiff" field inside $project, so that it will be shown in the final display:

    [ ]:
                    "date_min": {"$min": "$timestamp"},
    If we specify unit as `day`, it will show the difference between the dates:

    If we specify unit as day, it will show the difference between the dates:

    [ ]:
                    "date_min": {"$min": "$timestamp"},
    <font size="+1">Practice</font>

    Practice

    Try it yourself find the date difference for each measurement type at site 29 in Nairobi.

    [ ]:
    pp.pprint(list(result))
    We can do more with the date data using `$dateTrunc`, which truncates datetime data. We need to specify the datetime data, which can be a `Date`, a `Timestamp`, or an `ObjectID`. Then we need to specify the `unit` (year, month, day, hour, minute, second) and `binSize` (numerical variable defining the size of the truncation). Let's check the example below, where we group data by the month using `$dateTrunc` and then count how many observations there are for each month.

    We can do more with the date data using $dateTrunc, which truncates datetime data. We need to specify the datetime data, which can be a Date, a Timestamp, or an ObjectID. Then we need to specify the unit (year, month, day, hour, minute, second) and binSize (numerical variable defining the size of the truncation). Let's check the example below, where we group data by the month using $dateTrunc and then count how many observations there are for each month.

    [ ]:
                                "date": "$timestamp",
    <font size="+1">Practice</font>

    Practice

    Try it yourself! Truncate date by week and count at site 29 in Nairobi.

    [ ]:
    pp.pprint(list(result))
    ## Finishing Up

    Finishing Up¶

    So far, we've connected to a server, accessed that server with a client, found the collection we were looking for within a database, and explored that collection in all sorts of different ways. Now it's time to get the data we'll actually need to build a model, and store that in a way we'll be able to use.

    Let's use find to retrieve the PM 2.5 data from site 3. And, since we don't need any of the metadata to build our model, let's strip that out using the projection argument. In this case, we're telling the collection that we only want to see "timestamp" and "P2". Keep in mind that we limited the number of records we'll get back to 3 when we defined result above.

    [ ]:
        # `projection` limits the kinds of data we'll get back.
    Finally, we'll use pandas to read the extracted data into a DataFrame, making sure to set `timestamp` as the index:

    Finally, we'll use pandas to read the extracted data into a DataFrame, making sure to set timestamp as the index:

    [ ]:
    df = pd.DataFrame(result).set_index("timestamp")
    <font size="+1">Practice</font>

    Practice

    Try it yourself! Retrieve the PM 2.5 data from site 29 in Nairobi and strip out the metadata to create a DataFrame that shows only timestamp and P2. Print the result.

    [ ]:
    result = ...
    # References & Further Reading

    References & Further Reading¶

    • Further reading about servers and clients
    • Definitions from the MongoDB documentation
    • Information on Iterators
    • MongoDB documentation in Aggregation
    ---

    Copyright 2022 WorldQuant University. This content is licensed solely for personal use. Redistribution or publication of this material is strictly prohibited.

    ​x
     

    Usage Guidelines

    This lesson is part of the DS Lab core curriculum. For that reason, this notebook can only be used on your WQU virtual machine.

    This means:

    • ⓧ No downloading this notebook.
    • ⓧ No re-sharing of this notebook with friends or colleagues.
    • ⓧ No downloading the embedded videos in this notebook.
    • ⓧ No re-sharing embedded videos with friends or colleagues.
    • ⓧ No adding this notebook to public or private repositories.
    • ⓧ No uploading this notebook (or screenshots of it) to other websites, including websites for study resources.

    <font size="+3"><strong>Machine Learning: Classification</strong></font>

    Machine Learning: Classification

    [ ]:
    warnings.simplefilter(action="ignore", category=FutureWarning)
    # Data Preprocessing

    Data Preprocessing¶

    For the examples here, we'll look at buildings in the Ramechhap district of Nepal. (In our SQLite database, Ramechhap has the `district_id` of `1`.) Run the wrangle function below to connect to the SQLite database load the data into the DataFrame `df`.

    For the examples here, we'll look at buildings in the Ramechhap district of Nepal. (In our SQLite database, Ramechhap has the district_id of 1.) Run the wrangle function below to connect to the SQLite database load the data into the DataFrame df.

    [ ]:
        df["damage_grade"] = pd.to_numeric(df["damage_grade"], errors="coerce")
    [ ]:
    df.head()
    # Data Segregation

    Data Segregation¶

    ## Training Sets

    Training Sets¶

    ### Randomized Train-Test split

    Randomized Train-Test split¶

    **Splitting a dataset** into different sets is an important part of the model development process. The initial dataset is typically split into **two** (**training** and **testing**) or **three** (**training**, **validation**, and **testing**) datasets. This helps ensure that the model can generalize. Usually, more data is used for training than for validation or testing. If splitting into two datasets, a good rule of thumb is to split your data randomly into a ratio of **80:20** training:testing. If splitting into three datasets, splitting the data into a ratio of **70:20:10** (training:validation:testing) is commonly used. 

    Splitting a dataset into different sets is an important part of the model development process. The initial dataset is typically split into two (training and testing) or three (training, validation, and testing) datasets. This helps ensure that the model can generalize. Usually, more data is used for training than for validation or testing. If splitting into two datasets, a good rule of thumb is to split your data randomly into a ratio of 80:20 training:testing. If splitting into three datasets, splitting the data into a ratio of 70:20:10 (training:validation:testing) is commonly used.

    Validation datasets are usually used to tune model hyperparameters. A hyperparameter is a model setting that can't be learned during model training and must be explicitly set. In contrast, a model parameter can be learned. An example of a hyperparameter is the depth of a decision tree . An example of a model parameter includes a coefficient of a variable from linear regression.

    In order to split our data, we'll be using the train_test_split function from scikit-learn. We'll begin by splitting our data into a training and testing set. Next, we'll apply the train_test_split function to our testing set to generate our validation dataset and new testing dataset.

    We will create a feature matrix X and target vector y. The target is "severe_damage".

    [ ]:
    target = "severe_damage"
    Drop the target from the DataFrame and save the results into a X. Save the target column into y. 

    Drop the target from the DataFrame and save the results into a X. Save the target column into y.

    [ ]:
    X = ...
    Finally, we will split our dataset into a training and test set using the `train_test_split` function from `scikit-learn`. 

    Finally, we will split our dataset into a training and test set using the train_test_split function from scikit-learn.

    [ ]:
    X_train, X_test, y_train, y_test = train_test_split(
    ## Validation Set

    Validation Set¶

    <font size="+1">Practice: Perform a randomized split using scikit-learn</font>

    Practice: Perform a randomized split using scikit-learn

    Try it yourself! Use train_test_split to divide the training data (X_train and y_train) into training and validation sets using the same randomized train-test split function used previously. The validation data will be 20% of the previously constructed training data. Don't forget to set a random_state.

    [ ]:
    X_train, X_val, y_train, y_val = ...
    # Key Concepts

    Key Concepts¶

    ## Majority and Minority Classes

    Majority and Minority Classes¶

    The majority class refers to whatever category in a binary target occurs most frequently, and the minority class refers to whatever category in a binary target occurs less frequently. Let's use the `value_counts` method to plot the relative frequency of the two plots with a bar chart.  

    The majority class refers to whatever category in a binary target occurs most frequently, and the minority class refers to whatever category in a binary target occurs less frequently. Let's use the value_counts method to plot the relative frequency of the two plots with a bar chart.

    [ ]:
        kind="bar", xlabel="Group", ylabel="Relative Frequency"
    Since the category 1 (`severe_damage` = True) occurs most frequently, this is the majority class. 

    Since the category 1 (severe_damage = True) occurs most frequently, this is the majority class.

    ## Positive and Negative Classes

    Positive and Negative Classes¶

    **Positive class** and **negative class** are the two possible labels for binary classification problems. For example, if we are classifying whether an email is spam or not, we can designate "spam" as the positive class and "not spam" as the negative class. For the example in the project, we have "bankrupt" as the positive class and "not bankrupt" as the negative class. Conventionally, we use `0` or `False` to represent negative class, and `1` or `True` to represent positive class.

    Positive class and negative class are the two possible labels for binary classification problems. For example, if we are classifying whether an email is spam or not, we can designate "spam" as the positive class and "not spam" as the negative class. For the example in the project, we have "bankrupt" as the positive class and "not bankrupt" as the negative class. Conventionally, we use 0 or False to represent negative class, and 1 or True to represent positive class.

    # Classification with Logistic Regression

    Classification with Logistic Regression¶

    ## Logistic Regression

    Logistic Regression¶

    The logistic regression model is the classifier version of linear regression. It will predict probability values that can be used to assign class labels. The model works by taking the output of a linear regression model and feeding it into a sigmoid or logistic function.

    Why transform a linear model this way? Linear regression models are great for regression problems because they can give you predictions that range from negative infinity to positive infinity. However, the sigmoid function bounds predictions between 0 and 1, which we then treat as a probability. This allows us to use the model for classification problems.

    An example of the sigmoid function is shown below.

    [ ]:
    x = np.linspace(-10, 10, 100)
    The following video summarizes the math behind logistic regression:

    The following video summarizes the math behind logistic regression:

    [ ]:
    YouTubeVideo("yIYKR4sgzI8")
    You can add the logistic regression as a named step in a model pipeline like below:

    You can add the logistic regression as a named step in a model pipeline like below:

    [ ]:
    model = make_pipeline(OneHotEncoder(), LogisticRegression(max_iter=1000))
    ## High-cardinality Features

    High-cardinality Features¶

    Cardinality refers to the number of unique values in a categorical variable. High cardinality means the categorical features have a large number of unique values. These features often don't work well with either one hot encoding or ordinal encoding. There is no exact number of unique values that makes a feature high-cardinality, but if the value of the categorical feature is unique for almost all observations, it can usually be dropped. You can see the number of unique values in a variable by using the `value_counts` method. For example, to check the number of unique values in the `roof_type` column:

    Cardinality refers to the number of unique values in a categorical variable. High cardinality means the categorical features have a large number of unique values. These features often don't work well with either one hot encoding or ordinal encoding. There is no exact number of unique values that makes a feature high-cardinality, but if the value of the categorical feature is unique for almost all observations, it can usually be dropped. You can see the number of unique values in a variable by using the value_counts method. For example, to check the number of unique values in the roof_type column:

    [ ]:
    df["roof_type"].value_counts()
    There are only three unique values, so we will leave the column in the DataFrame. 

    There are only three unique values, so we will leave the column in the DataFrame.

    Practice

    Try it yourself! Use value_counts to check the number of unique values in the building_id column. Remove the column in the wrangle function if it has a large number of unique values.

    [ ]:
    # X_train["building_id"].value_counts()
    # Classification with Tree-based Models

    Classification with Tree-based Models¶

    ## Decision Trees

    Decision Trees¶

    Decision trees are a general class of machine learning models that are used for both classification and regression. The model resemble a tree, complete with branches and leaves. The model is essentially a series of questions with "yes" or "no" answers. The decision tree starts by checking whatever condition does the best job at correctly separating the data into the two classes in the binary target. It then progressively checks more conditions until it can predict an observation's label. They are popular because they are more flexible than linear models and intuitive in a way that makes them easy to explain to stakeholders who are not familiar with data science.  

    Decision trees are a general class of machine learning models that are used for both classification and regression. The model resemble a tree, complete with branches and leaves. The model is essentially a series of questions with "yes" or "no" answers. The decision tree starts by checking whatever condition does the best job at correctly separating the data into the two classes in the binary target. It then progressively checks more conditions until it can predict an observation's label. They are popular because they are more flexible than linear models and intuitive in a way that makes them easy to explain to stakeholders who are not familiar with data science.

    Decision trees pros and cons:

    Decision trees pros and cons:

    Pros Cons
    can be used for classification and regression generalization: they are prone to overfitting
    handles both numerical and categorical data robustness: small variations in data can result in a different tree
    models nonlinear relationships between the features and target class imbalance: if one class is much larger than the other, the tree may be unbalanced
    The following video summarizes a decision tree:

    The following video summarizes a decision tree:

    [ ]:
    YouTubeVideo("7VeUPuFGJHk")
    We will fit a decision tree to the training data, using an ordinal encoder to encode the categorical features:

    We will fit a decision tree to the training data, using an ordinal encoder to encode the categorical features:

    [ ]:
        OrdinalEncoder(), DecisionTreeClassifier(max_depth=6, random_state=42)
    ## Prediction

    Prediction¶

    ## Probability Estimates

    Probability Estimates¶

    Sometimes a model makes the same prediction for the target of two observations, but is more certain about one prediction. This is the difference between the prediction and the prediction's associated probability. 

    Sometimes a model makes the same prediction for the target of two observations, but is more certain about one prediction. This is the difference between the prediction and the prediction's associated probability.

    The predict method predicts the target of an unlabeled observation. The predict_proba outputs the probability that an unlabeled observation belongs to one of two classes in the target. Both methods work similarly. They each are run on the fitted model and take a set of features as their input. For example, if we want to see the associated predictions if we used the X_train as an input:

    [ ]:
    model.predict(X_train)
    And if we wanted to see the associated probabilities for these predictions:

    And if we wanted to see the associated probabilities for these predictions:

    [ ]:
    model.predict_proba(X_train)
    Note that there are two probability estimates for each observation, one for the likelihood of each class in the target. The second probability is the likelihood that the unknown observation belongs to the class equal to 1 and the first probability is the likelihood that the unknown observation belongs to the class equal to 0. Whichever class's probability is higher is the predicted class from `predict`. 

    Note that there are two probability estimates for each observation, one for the likelihood of each class in the target. The second probability is the likelihood that the unknown observation belongs to the class equal to 1 and the first probability is the likelihood that the unknown observation belongs to the class equal to 0. Whichever class's probability is higher is the predicted class from predict.

    <font size="+1">Practice: Generate probability estimates using a trained model in scikit-learn</font>

    Practice: Generate probability estimates using a trained model in scikit-learn

    Try it yourself! Use predict_proba to generate probability estimates for the observations in X_test.

    [ ]:
    model.predict_proba(X_test)
    ## Evaluation

    Evaluation¶

    ### Calculating Accuracy Score

    Calculating Accuracy Score¶

    A natural choice for a metric for classification is accuracy. Accuracy is equal to the number of observations you correctly classified over all observations. For example, if your model properly identified 77 out of 100 images, you have an accuracy of 77%. Accuracy is an easy metric to both understand and calculate. Mathematically, it is simply

    A natural choice for a metric for classification is accuracy. Accuracy is equal to the number of observations you correctly classified over all observations. For example, if your model properly identified 77 out of 100 images, you have an accuracy of 77%. Accuracy is an easy metric to both understand and calculate. Mathematically, it is simply

    number of correct observationsnumber of observations.number of correct observationsnumber of observations.

    Model accuracy can be calculated using the accuracy_score function. The function requires two arguments, the true labels and the predicted labels. For example, if we want to calculate the model accuracy score on the training data:

    [ ]:
    acc_train = accuracy_score(y_train, model.predict(X_train))
    <font size="+1">Practice: Calculate the accuracy score for a model in scikit-learn</font>

    Practice: Calculate the accuracy score for a model in scikit-learn

    Try it yourself! Calculate the model's accuracy on the validation data:

    [ ]:
    print("Validation Accuracy:", round(acc_val, 2))
    ### Baseline Accuracy Score

    Baseline Accuracy Score¶

    How do you know whether or not the accuracy score you calculated for your model is good? A baseline accuracy score for the model can be used to compare your model accuracy results against. A common baseline is to use the percentage that the majority class shows up in the training data. This would be your accuracy if you simply predicted the majority class for all observations. If the model is not beating this baseline, that suggests that the features are not adding any valuable information to classify your observations.

    We can use the value_counts method with the normalize = True argument to calculate the baseline accuracy:

    [ ]:
    acc_baseline = y_train.value_counts(normalize=True).max()
    ### Confusion Matrix

    Confusion Matrix¶

    Accuracy score may not provide enough information to assess how a model is performing because it only gives us an overall score. Also, imbalanced data can lead to a high accuracy score even when a model isn't particularly useful. If we want to know what fraction of all positive predictions were correct and what fraction of positive observations did we identify, we can use a **confusion matrix**.

    Accuracy score may not provide enough information to assess how a model is performing because it only gives us an overall score. Also, imbalanced data can lead to a high accuracy score even when a model isn't particularly useful. If we want to know what fraction of all positive predictions were correct and what fraction of positive observations did we identify, we can use a confusion matrix.

    A confusion matrix is a table summarizing the performance of the model by enumerating true and false positives and the true and false negatives.

    Positive Observation Negative Observation
    Positive Prediction True Positive (TP) False Positive (FP)
    Negative Prediction False Negative (FN) True Negative (TN)
    Refer to this video for more details in confusion matrix:

    Refer to this video for more details in confusion matrix:

    [ ]:
    YouTubeVideo("_cpiuMuFj3U")
    Here is the code to get the confusion matrix in the training set:

    Here is the code to get the confusion matrix in the training set:

    [ ]:
    cm = confusion_matrix(y_train, model.predict(X_train))
    You can also use the heatmap to better visualize confusion matrix using `ConfusionMatrixDisplay`:

    You can also use the heatmap to better visualize confusion matrix using ConfusionMatrixDisplay:

    [ ]:
    disp = ConfusionMatrixDisplay(confusion_matrix=cm, display_labels=model.classes_)
    <font size="+1">Practice</font>

    Practice

    Get confusion matrix for the validation set and display with ConfusionMatrixDisplay:

    [ ]:
    disp.plot()
    ### Precision Score

    Precision Score¶

    Depending on the context of the problem, instead of knowing model performances in both classes, sometimes we are more interested in the results in positive class. That's when we use **precision**. Precision is the fraction of true positives over all positive predictions. It is a measure of how "precise" our model is with regard to labeling observations as positive. 

    Depending on the context of the problem, instead of knowing model performances in both classes, sometimes we are more interested in the results in positive class. That's when we use precision. Precision is the fraction of true positives over all positive predictions. It is a measure of how "precise" our model is with regard to labeling observations as positive.

    For example in Project 3, we try to predict whether a company will go bankrupt, with "bankrupt" as the positive class. Out of all positive predictions made by the model, some companies actually went bankrupt (True Positive TP), while others didn't (False Positive FP). Precision measures how many times model predicted positives (TP+FP) correctly (TP). The equation for precision is:

    precision=TP𝑇𝑃+𝐹𝑃precision=TPTP+FP

    Using the data and model above, we can get a precision score using the code below:

    Using the data and model above, we can get a precision score using the code below:

    [ ]:
    precision = precision_score(y_train, model.predict(X_train))
    <font size="+1">Practice</font>

    Practice

    Get precision for the validation set

    [ ]:
    print(f"Validation Set Precision is {round(precision_val, 2)}")
    ### Recall Score

    Recall Score¶

    What if we care more about the model performance in the negative class? In this case, we need to calculate **recall**. Recall the fraction of true positives over all positive observations. It is a measure of our model's ability to "catch" and properly label observations that are positive. 

    What if we care more about the model performance in the negative class? In this case, we need to calculate recall. Recall the fraction of true positives over all positive observations. It is a measure of our model's ability to "catch" and properly label observations that are positive.

    Let's return to the Poland bankruptcy example. Of all the companies that actually went bankrupt (TP+FN), how many companies did out model predict as going bankrupt (TP)? That's what recall measures. The equation to calculate recall is:

    Recall=TPTP+FN.Recall=TPTP+FN.

    Here is the code to calculate recall:

    Here is the code to calculate recall:

    [ ]:
    recall = recall_score(y_train, model.predict(X_train))
    <font size="+1">Practice</font>

    Practice

    Get precision for the validation set

    [ ]:
    print(f"Validation Set Precision is {round(recall_val, 2)}")
    ### Classification Report

    Classification Report¶

    We can also use a **classification report** to look at the whole picture of the classification model performances. A classification report includes precision, recall, **F1 score** and **support**. We already know the first two, but F1 score is the harmonic mean of precision and recall, it equation is:

    We can also use a classification report to look at the whole picture of the classification model performances. A classification report includes precision, recall, F1 score and support. We already know the first two, but F1 score is the harmonic mean of precision and recall, it equation is:

    F1=2(precision⋅recallprecision+recall)F1=2(precision⋅recallprecision+recall)

    Support number of observations for each class, thus it is useful to understand whether the data is imbalanced or not.

    [ ]:
    print(classification_report(y_train, model.predict(X_train)))
    Note in the last two rows, we have the macro and the weighted average,. Macro average is the arithmetic average of a metric between the two classes:

    Note in the last two rows, we have the macro and the weighted average,. Macro average is the arithmetic average of a metric between the two classes:

    0.65=0.70+0.6020.65=0.70+0.602

    The weighted average is calculated as:

    ∑(metric of interest⋅weight)∑(weights)∑(metric of interest⋅weight)∑(weights)

    Here the weights are the number of observation for each class.

    Here you may notice there are two rows of metrics. If you refer back to what we calculated previous on precision and recall, the second row actually align with what we found. That's because we usually define class one as the **positive class**, thus we are referring class 1's metric performance as the true precision and recall value.

    Here you may notice there are two rows of metrics. If you refer back to what we calculated previous on precision and recall, the second row actually align with what we found. That's because we usually define class one as the positive class, thus we are referring class 1's metric performance as the true precision and recall value.

    <font size="+1">Practice</font>

    Practice

    Get classification report for the validation set

    [ ]:
    print(...)
    ## Communication

    Communication¶

    ### Plotting a Decision Tree

    Plotting a Decision Tree¶

    The `plot_tree` function can be used to a plot a decision tree. The visualization is fit to the size of the axis set with `matplotlib`. Use the `figsize` argument of `plt.subplots` to control the size of the tree.

    The plot_tree function can be used to a plot a decision tree. The visualization is fit to the size of the axis set with matplotlib. Use the figsize argument of plt.subplots to control the size of the tree.

    We'll demonstrate how to use the `plot_tree` function to graphically display a decision tree:

    We'll demonstrate how to use the plot_tree function to graphically display a decision tree:

    [ ]:
        proportion=True,  # Display proportion of classes in leaf
    <font size="+1">Practice: Plot a decision tree using `scikit-learn`</font>

    Practice: Plot a decision tree using scikit-learn

    Try it yourself! Use plot_tree to plot the decision tree from the model object, modifying the parameters of the tree to only display the first 3 levels and to not display the proportion of classes in a leaf.

    [ ]:
    fig, ax = plt.subplots(figsize=(25, 12))
    The feature names and importance of features can be extracted from the column names in your training set. For the `importances`, you can access the [`feature_importances_`](https://scikit-learn.org/stable/modules/generated/sklearn.tree.DecisionTreeClassifier.html#sklearn.tree.DecisionTreeClassifier.feature_importances_) attribute of your model's `DecisionTreeClassifier`. 

    The feature names and importance of features can be extracted from the column names in your training set. For the importances, you can access the feature_importances_ attribute of your model's DecisionTreeClassifier.

    [ ]:
    importances = model.named_steps["decisiontreeclassifier"].feature_importances_
    The importance of a feature is based on how well the feature correctly classifies observations. In a decision tree, this is on average how much a feature reduces the impurity metric. The tree determines how to split based on an impurity function. The impurity function calculates how homogeneous observations are at a particular leaf node. Conditions that do a better job minimizing impurity are used to split first. The `sklearn.tree` algorithm uses the Gini impurity by default. The Gini impurity measures the probability of an incorrect classification in the model for each branch. It ranges from 0 to 1. 

    The importance of a feature is based on how well the feature correctly classifies observations. In a decision tree, this is on average how much a feature reduces the impurity metric. The tree determines how to split based on an impurity function. The impurity function calculates how homogeneous observations are at a particular leaf node. Conditions that do a better job minimizing impurity are used to split first. The sklearn.tree algorithm uses the Gini impurity by default. The Gini impurity measures the probability of an incorrect classification in the model for each branch. It ranges from 0 to 1.

    Let's create a bar chart to plot each feature with its corresponding importance. To build this bar chart, we'll create a pandas Series named feat_imp, where the index is features and the values are your importances. The Series should be sorted from smallest to largest importance so that the bar chart is also in order. 

    Let's create a bar chart to plot each feature with its corresponding importance. To build this bar chart, we'll create a pandas Series named feat_imp, where the index is features and the values are your importances. The Series should be sorted from smallest to largest importance so that the bar chart is also in order.

    [ ]:
    feat_imp = pd.Series(importances, index=features).sort_values()
    Next, we'll use the series to build a bar chart:

    Next, we'll use the series to build a bar chart:

    [ ]:
    plt.xlabel("Gini Importance");
    # Classification with Ensemble Models

    Classification with Ensemble Models¶

    Ensemble models are machine learning models that use more than one predictor to arrive at a prediction. A group of predictors form an _ensemble_. In general, ensemble models perform better than using a single predictor. There are three types of ensemble models: **bagging**, **boosting**, and **blending**. Of the three, decision trees are commonly used to construct bagging and boosting models.

    Ensemble models are machine learning models that use more than one predictor to arrive at a prediction. A group of predictors form an ensemble. In general, ensemble models perform better than using a single predictor. There are three types of ensemble models: bagging, boosting, and blending. Of the three, decision trees are commonly used to construct bagging and boosting models.

    ## Random Forest

    Random Forest¶

    The performance of a single decision tree will be limited. Instead of relying on one tree, a better approach is to aggregate the predictions of multiple trees. On average, aggregation will perform better than a single predictor. You can envision the aggregation as mimicking the idea of "wisdom of the crowd." We call a tree based model that aggregates the predictions of multiple trees a **random forest**.

    The performance of a single decision tree will be limited. Instead of relying on one tree, a better approach is to aggregate the predictions of multiple trees. On average, aggregation will perform better than a single predictor. You can envision the aggregation as mimicking the idea of "wisdom of the crowd." We call a tree based model that aggregates the predictions of multiple trees a random forest.

    In order for a random forest to be effective, the model needs a diverse collection of trees. There should be variations in the chosen thresholds for splitting and the number of nodes and branches. There is no point in aggregating the predicted results if all the trees are nearly identical and produce the same result. There is no "wisdom of the crowd" if everyone thinks alike. To achieve a diverse set of trees, we need to:

    1. Train each tree in the forest using a different subset of the training set.
    2. Only consider a subset of features when deciding how to split the nodes.

    On the first point, we would ideally generate a new training set for each tree. However, oftentimes it's too difficult or expensive to collect more data, so we have to make do with what we have. Bootstrapping is a general statistical technique to generate "new" data sets with a single set by random sampling with replacement. Sampling with replacement allows for a data point to be sampled more than once.

    Typically, when training the standard decision tree model, the algorithm will consider all features in deciding the node split. Considering only a subset of your features ensures that your trees do not resemble each other. If the algorithm had considered all features, a dominant feature would be continuously chosen for node splits.

    The hyperparameters available for random forests include those of decision tress with some additions.

    Hyperparameter Description
    n_estimators The number of trees in the forest
    max_samples If bootstrap is True, the number of samples to draw from X to train each base estimator
    max_features The number of features to consider when looking for the best split
    n_jobs The number of jobs to run in parallel when fitting and predicting
    warm_start If set to True, reuse the trained tree from a prior fitting and just train the additional trees

    Since the random forest is based on idea of bootstrapping and aggregating the results, it is referred to as a bagging ensemble model.

    ## Gradient Boosting Trees

    Gradient Boosting Trees¶

    Gradient boosting trees is another ensemble model. It uses a collection of tree models arranged in a sequence. Here, the model is built stage-wise; each additional tree aims to correct the previous tree's incorrect. 

    Gradient boosting trees is another ensemble model. It uses a collection of tree models arranged in a sequence. Here, the model is built stage-wise; each additional tree aims to correct the previous tree's incorrect.

    Where does the name gradient in gradient boosting trees come from? Gradient descent is a minimization algorithm that updates/improves the current answer by taking a step in the direction of minimizing the loss function. This is the same as the gradient boosting trees algorithm as it adds trees to minimize loss/improve model performance. The term boosting refers to the algorithm's ability to combine multiple weak models in sequence to form a stronger model.

    Gradient boosting trees have a similar set of hyperparameters as random forests but with some key additions.

    Gradient boosting trees have a similar set of hyperparameters as random forests but with some key additions.

    Hyperparameter Description
    learning_rate Multiplicative factor of the tree's contribution to the model.
    subsample Fraction of the training data to use when fitting the trees.
    The learning rate determines how much each tree affect the final outcome and is very important in model convergence. Thus it should be considered during hyperparameter tuning to improve model performance.

    The learning rate determines how much each tree affect the final outcome and is very important in model convergence. Thus it should be considered during hyperparameter tuning to improve model performance.

    # Hyperparameter Tuning

    Hyperparameter Tuning¶

    When we defined our decision tree estimator, we chose how many layers the tree would have using the `max_depth` argument. More generally, when we instantiate any estimator, we can pass keyword arguments that will dictate its structure. The decision tree regressor accepts [12 different keyword arguments](http://scikit-learn.org/stable/modules/generated/sklearn.tree.DecisionTreeRegressor.html#sklearn.tree.DecisionTreeRegressor). These arguments are called **hyperparameters**. This is in contrast to **parameters**, which are the numbers that our model uses to predict labels based on features. Hyperparameters are decided before training and dictate the model's structure. Parameters are optimized during training. Basically all models have hyperparameters. Even a simple linear regressor has a hyperparameter: `fit_intercept`.

    When we defined our decision tree estimator, we chose how many layers the tree would have using the max_depth argument. More generally, when we instantiate any estimator, we can pass keyword arguments that will dictate its structure. The decision tree regressor accepts 12 different keyword arguments. These arguments are called hyperparameters. This is in contrast to parameters, which are the numbers that our model uses to predict labels based on features. Hyperparameters are decided before training and dictate the model's structure. Parameters are optimized during training. Basically all models have hyperparameters. Even a simple linear regressor has a hyperparameter: fit_intercept.

    Since changing a hyperparameters will change the structure of the model, we should think of choosing hyperparameters as part of the model building process. We can usually use cross validation combined with grid search when looking for the best model with the right hyperparameters. This process is called hyperparameter tuning.

    ## Cross-Validation

    Cross-Validation¶

    When trying out different hyperparameter settings for estimators (such as the `max_depth` for a random forest), there's a risk in using the test set to evaluate these settings. What happens is that your knowledge about the test set can “leak” into the model, and performance metrics no longer reflect the model's ability to generalize. 

    When trying out different hyperparameter settings for estimators (such as the max_depth for a random forest), there's a risk in using the test set to evaluate these settings. What happens is that your knowledge about the test set can “leak” into the model, and performance metrics no longer reflect the model's ability to generalize.

    The generalization problem can be solved adding an extra set called validation set. In this case, we train the model with the training set, then evaluate different hyperparameters using the validation set. If the model is performing well in both sets, finally we will evaluate the model on the test set.

    But there's a drawback to this strategy. The potential issue we may face dividing data into three sets is that we will reduce the number of samples available to fit and train the model. In addition, the model results will change with respect to difference choices of training and validation portions.

    The solution here is to use cross validation (CV for short). In this case, we will still use a test set, but a validation set is no longer needed. k-fold CV is the most used cross validation method(http://scikit-learn.org/stable/modules/cross_validation.html#k-fold). The algorithm divides the training set into 𝑘k small folds. For each fold 𝑘k, we:

    1. Train the model using all the folds but one (i.e. 𝑘−1k−1 folds) as training data;

    2. Validate the model using the remaining fold as if it were test data, and store the performance metric;

    This approach makes the best use of all the data we are given, so it's particularly useful when the sample size is small.

    Here is the code for conducting a 5-fold cross-validation and reporting the accuracy score for each fold.

    Here is the code for conducting a 5-fold cross-validation and reporting the accuracy score for each fold.

    [ ]:
    clf = make_pipeline(OneHotEncoder(), DecisionTreeClassifier())
    [ ]:
        f"{round(scores.mean(),2)} accuracy with a standard deviation of {round(scores.std(),2)}"
    <font size="+1">Practice</font>

    Practice

    Perform 2-fold Cross ValidationWQU WorldQuant University Applied Data Science Lab QQQQ

    [ ]:
    scores = ...
    ## Grid Search

    Grid Search¶

    Another a useful tool for comparing different hyperparameter values is `GridSearchCV`. There are two ideas behind `GridSearchCV`: first we split the data using k-fold cross-validation, and then we train and evaluate models with different hyperparameter settings selected from a grid of possible combinations.

    Another a useful tool for comparing different hyperparameter values is GridSearchCV. There are two ideas behind GridSearchCV: first we split the data using k-fold cross-validation, and then we train and evaluate models with different hyperparameter settings selected from a grid of possible combinations.

    First, we need to define the hyperparameters we want to tune, and tuning in what range. Here we are using an example of searching the best value for the `max_depth` in decision tree model. Since we will be building a pipeline including a transformer and estimator, we need to specify `max_depth` comes from the estimator `decisiontreeclassifier`.

    First, we need to define the hyperparameters we want to tune, and tuning in what range. Here we are using an example of searching the best value for the max_depth in decision tree model. Since we will be building a pipeline including a transformer and estimator, we need to specify max_depth comes from the estimator decisiontreeclassifier.

    [ ]:
    params = {"decisiontreeclassifier__max_depth": range(1, 15)}
    The we define the pipeline and the model with `GridSearchCV`.

    The we define the pipeline and the model with GridSearchCV.

    [ ]:
    clf = make_pipeline(OneHotEncoder(), DecisionTreeClassifier())
    Lastly fit the model:

    Lastly fit the model:

    [ ]:
    model.fit(X_train, y_train)
    We can check the best parameters once the fitting process finished:

    We can check the best parameters once the fitting process finished:

    [ ]:
    model.best_params_
    <font size="+1">Practice</font>

    Practice

    Perform GridSearchCV on both max_depth and criterion for the validation set

    [ ]:
    # Check the best hyperparameters
    # References & Further Reading

    References & Further Reading¶

    • scikit-learn documentation on decision tree classifier model object
    • scikit-learn documentation on decision tree plot
    • scikit-learn documentation on decision tree math
    • scikit-learn confusion matrix
    • scikit-learn precision score
    • scikit-learn recall score
    • scikit-learn classification report
    • YoutubeConfusion Matrix
    • scikit-learnCross Validation
    • scikit-learnk-fold Cross Validation
    ---

    Copyright 2022 WorldQuant University. This content is licensed solely for personal use. Redistribution or publication of this material is strictly prohibited.

    ​x
     

    Usage Guidelines

    This lesson is part of the DS Lab core curriculum. For that reason, this notebook can only be used on your WQU virtual machine.

    This means:

    • ⓧ No downloading this notebook.
    • ⓧ No re-sharing of this notebook with friends or colleagues.
    • ⓧ No downloading the embedded videos in this notebook.
    • ⓧ No re-sharing embedded videos with friends or colleagues.
    • ⓧ No adding this notebook to public or private repositories.
    • ⓧ No uploading this notebook (or screenshots of it) to other websites, including websites for study resources.

    <font size="+3"><strong>Machine Learning: Core Concepts</strong></font>

    Machine Learning: Core Concepts

    # Model Types

    Model Types¶

    **Linear regression** is a way to predict the value of some a target variable by fitting a line that best describes the relationship between **Big X** and **little y** for the values we already have. If you remember `y = mx + b` from Algebra, it's the same thing; the `y` is the intercept, and the `b` is the beta coefficient. The beta coefficient tells us what change we can expect to see in `X` for every one-unit increase in `y`. If that doesn't seem familiar to you, don't worry about it; we'll give you everything you need to know.

    Linear regression is a way to predict the value of some a target variable by fitting a line that best describes the relationship between Big X and little y for the values we already have. If you remember y = mx + b from Algebra, it's the same thing; the y is the intercept, and the b is the beta coefficient. The beta coefficient tells us what change we can expect to see in X for every one-unit increase in y. If that doesn't seem familiar to you, don't worry about it; we'll give you everything you need to know.

    # Statistical Concepts 

    Statistical Concepts¶

    ## Cost Functions

    Cost Functions¶

    When we train a model, we're solving an optimization problem. We provide training data to an algorithm and tell it to find the model or model parameters that best fit the data. But how can the algorithm judge what the "best" fit is? What criteria should it use?

    A cost function (sometimes also called a loss or error function) is a mathematical formula that provides the score by which the algorithm will determine the best fit. Generally, the the goal is to minimize the cost function and get the lowest score. For linear models, these functions measure distance, and the model tries to to get the closest fit to the data. For tree-based models, they measure impurity, and the model tries to get the most terminal nodes.

    ## Residuals

    Residuals¶

    When we perform any type of regression analysis, we end up with a line of best fit. Because our data comes from the real world, it tends to be a little bit messy, so the data points usually don’t fall exactly on this line. Most of the time, they’re are scattered around it, and a residual is the vertical distance between each individual data point and the regression line. Each data point has only one residual which can be positive if it’s above the regression line, negative if it’s below the regression line, or zero if the line passes directly through the point. Think of it like this: the model describes theoretical line. That line doesn't really exist outside the model. The residuals, however, are true values; they represent the actual data that came from real observations.

    ## Performance Metrics

    Performance Metrics¶

    In statistics, an error is the difference between a measurement and reality. There may not be any difference at all, but there's usually something not quite right, and we need to account for that in our model. To do that, we need to figure out the mean absolute error (MAE). Absolute error is the error in a single measurement, and mean absolute error is the average error over the course of several measurements.

    Imagine that you're buying some bananas. The store charges for fruit based on weight, so you put your bananas on a scale before you head off to pay for them. The scale says they weigh 1.2 kilos, but your innate sense of weight tells you that they actually weight 0.9 kilos. The absolute error in that measurement would be 0.3 kilos. It can go the other way too: maybe you know the bananas weight 1.2 kilos, but the scale says they were 0.9 kilos. In that case, the absolute error would still be 0.3 kilos, because even though the numerical difference is -0.3, absolute values are always positive; all you have to do is disregard the - sign.

    Let's keep going: you're sure the bananas don't weight 1.2 kilos, so you weigh them again. This time, the scale says 1.0 kilos. That's still wrong, so you weigh the bananas a third time, and now the scale says 2.3 kilos. Since the actual weight of your bananas hasn't changed, you now have a set of three absolute errors: 0.3, 0.1, and 1.4. If we average those errors together, we get 0.6, which is the mean absolute error for your banana data.

    # Data Concepts

    Data Concepts¶

    ## Leakage

    Leakage¶

    Leakage is the use of data in training your model that would not be typically be available when making predictions. For example, suppose we want to predict property prices in USD but include property prices in Mexican Pesos in our model. If we assume a fixed exchange rate or a nearly constant exchange rate, then our model will have a low error on the training data, but this will not be reflective of its performance on real world data.

    ## Imputation

    Imputation¶

    Datasets are often incomplete or missing entries. If the dataset is large and the missing entries are few, then the missing entries aren't all that important. But sometimes, it might be useful to include data with missing entries by finding a way to impute the missing entries in a row or column of a DataFrame. For example, you might use extrapolation when the data points have a pattern, or you might approximate the missing values by mean values.

    ## Generalization

    Generalization¶

    Notice that we tested the model with a dataset that's different from the one we used to train the model. Machine learning models are useful if they allow you to make predictions about data other than what you used to train your model. We call this concept generalization. By testing your model with different data than you used to train it, you're checking to see if your model can generalize. Most machine learning models do not generalize to all possible types of input data, so they should be used with care. On the other hand, machine learning models that don't generalize to make predictions for at least a restricted set of data aren't very useful.

    # Model Concepts

    Model Concepts¶

    ## Hyperparameters

    Hyperparameters¶

    When we instantiate an estimator, we can pass keyword arguments that will dictate its structure. These arguments are called **hyperparameters**. For example, when we defined our decision tree estimator, we chose how many layers the tree would have using the `max_depth` keyword. This is in contrast to **parameters**, which are the numbers that our model uses to make predictions based on features. Parameters are optimized during the training process based on data and input features. They keep changing during training to fit the data and only the best performed ones were selected.  Hyperparameters values are set before training begins and will not be changed during the training process. Pretty much all models have hyperparameters. Even a simple linear regressor has a hyperparameter: `fit_intercept`. Here are some common examples for Hyperparameters:

    When we instantiate an estimator, we can pass keyword arguments that will dictate its structure. These arguments are called hyperparameters. For example, when we defined our decision tree estimator, we chose how many layers the tree would have using the max_depth keyword. This is in contrast to parameters, which are the numbers that our model uses to make predictions based on features. Parameters are optimized during the training process based on data and input features. They keep changing during training to fit the data and only the best performed ones were selected. Hyperparameters values are set before training begins and will not be changed during the training process. Pretty much all models have hyperparameters. Even a simple linear regressor has a hyperparameter: fit_intercept. Here are some common examples for Hyperparameters:

    • The imputation strategy used for missing data.
    • The number of trees in a random forest model.
    • The number of jobs to run in parallel when fitting and predicting.
    # References and Further Reading

    References and Further Reading¶

    - [Parameters and Hyperparameters in Machine Learning and Deep Learning](https://towardsdatascience.com/parameters-and-hyperparameters-aa609601a9ac) 
    • Parameters and Hyperparameters in Machine Learning and Deep Learning
    ---

    Copyright 2022 WorldQuant University. This content is licensed solely for personal use. Redistribution or publication of this material is strictly prohibited. WQU WorldQuant University Applied Data Science Lab QQQQ

    ---

    Copyright 2022 WorldQuant University. This content is licensed solely for personal use. Redistribution or publication of this material is strictly prohibited.

    ​x
     

    Usage Guidelines

    This lesson is part of the DS Lab core curriculum. For that reason, this notebook can only be used on your WQU virtual machine.

    This means:

    • ⓧ No downloading this notebook.
    • ⓧ No re-sharing of this notebook with friends or colleagues.
    • ⓧ No downloading the embedded videos in this notebook.
    • ⓧ No re-sharing embedded videos with friends or colleagues.
    • ⓧ No adding this notebook to public or private repositories.
    • ⓧ No uploading this notebook (or screenshots of it) to other websites, including websites for study resources.

    <font size="+3"><strong>Visualizing Data: seaborn</strong></font>

    Visualizing Data: seaborn

    There are many ways to interact with data, and one of the most powerful modes of interaction is through **visualizations**. Visualizations show data graphically, and are useful for exploring, analyzing, and presenting datasets. We use four libraries for making visualizations: [pandas](../%40textbook/07-visualization-pandas.ipynb), [Matplotlib](../%40textbook/06-visualization-matplotlib.ipynb), [plotly express](../%40textbook/08-visualization-plotly.ipynb), and seaborn. In this section, we'll focus on using seaborn.

    There are many ways to interact with data, and one of the most powerful modes of interaction is through visualizations. Visualizations show data graphically, and are useful for exploring, analyzing, and presenting datasets. We use four libraries for making visualizations: pandas, Matplotlib, plotly express, and seaborn. In this section, we'll focus on using seaborn.

    # Scatter Plots

    Scatter Plots¶

    A **scatter plot** is a graph that uses dots to represent values for two different numeric variables. The position of each dot on the horizontal and vertical axis indicates values for an individual data point. Scatter plots are used to observe relationships between variables, and are especially useful if you're looking for **correlations**. 

    A scatter plot is a graph that uses dots to represent values for two different numeric variables. The position of each dot on the horizontal and vertical axis indicates values for an individual data point. Scatter plots are used to observe relationships between variables, and are especially useful if you're looking for correlations.

    In the following example, we will see some scatter plots based on the Mexico City real estate data. Specifically, we can use scatter plot to show how "price" and "surface_covered_in_m2" are correlated. First we need to read the data set and do a little cleaning.

    [1]:
    mexico_city1 = pd.read_csv("./data/mexico-city-real-estate-1.csv")
    [1]:
    operation property_type place_with_parent_names lat-lon price currency price_aprox_local_currency price_aprox_usd surface_total_in_m2 surface_covered_in_m2 price_per_m2 properati_url
    2 sell apartment |México|Distrito Federal|Cuauhtémoc| 19.41501,-99.175174 2700000.0 MXN 2748947.10 146154.51 61.0 61.0 44262.295082 http://cuauhtemoc.properati.com.mx/2pu_venta_a...
    3 sell apartment |México|Distrito Federal|Cuauhtémoc| 19.41501,-99.175174 6347000.0 MXN 6462061.92 343571.36 176.0 128.0 49585.937500 http://cuauhtemoc.properati.com.mx/2pv_venta_a...
    6 sell apartment |México|Distrito Federal|Miguel Hidalgo| 19.456564,-99.191724 670000.0 MXN 682146.11 36267.97 65.0 65.0 10307.692308 http://miguel-hidalgo-df.properati.com.mx/46h_...
    7 sell apartment |México|Distrito Federal|Gustavo A. Madero| 19.512787,-99.141393 1400000.0 MXN 1425379.97 75783.82 82.0 70.0 20000.000000 http://gustavo-a-madero.properati.com.mx/46p_v...
    8 sell house |México|Distrito Federal|Álvaro Obregón| 19.358776,-99.213557 6680000.0 MXN 6801098.67 361597.08 346.0 346.0 19306.358382 http://alvaro-obregon.properati.com.mx/46t_ven...
    Use seaborn to plot the scatter plot for `"price"` and `"surface_covered_in_m2"`:

    Use seaborn to plot the scatter plot for "price" and "surface_covered_in_m2":

    [2]:
    sns.scatterplot(data=mexico_city1, x="price", y="surface_covered_in_m2");
    There is a very useful argument in `scatterplot` called `hue`. By specifying a categorical column as `hue`, seaborn can create a scatter plot between two variables in different categories with different colors. Let's check the following example using `"property_type"`:

    There is a very useful argument in scatterplot called hue. By specifying a categorical column as hue, seaborn can create a scatter plot between two variables in different categories with different colors. Let's check the following example using "property_type":

    [3]:
        data=mexico_city1, x="price", y="surface_covered_in_m2", hue="property_type"
    <font size="+1">Practice</font>

    Practice

    Plot a scatter plot for "price" and "surface_total_in_m2" by "property_type" for "mexico-city-real-estate-1.csv":

    [ ]:
    ​x
     
    # Bar Charts

    Bar Charts¶

    A **bar chart** is a graph that shows all the values of a categorical variable in a dataset. They consist of an axis and a series of labeled horizontal or vertical bars. The bars depict frequencies of different values of a variable or simply the different values themselves. The numbers on the y-axis of a vertical bar chart or the x-axis of a horizontal bar chart are called the scale. 

    A bar chart is a graph that shows all the values of a categorical variable in a dataset. They consist of an axis and a series of labeled horizontal or vertical bars. The bars depict frequencies of different values of a variable or simply the different values themselves. The numbers on the y-axis of a vertical bar chart or the x-axis of a horizontal bar chart are called the scale.

    In the following example, we will see some bar plots based on the Mexico City real estate dataset. Specifically, we will count the number of observations in each borough and plot them. We first need to import the dataset and extract the borough and other location information from column "place_with_parent_names".

    [4]:
    ] = mexico_city1["place_with_parent_names"].str.split("|", 4, expand=True)
    /tmp/ipykernel_721/836102575.py:12: FutureWarning: In a future version of pandas all arguments of StringMethods.split except for the argument 'pat' will be keyword-only.
      ] = mexico_city1["place_with_parent_names"].str.split("|", 4, expand=True)
    
    [4]:
    operation property_type place_with_parent_names lat-lon price currency price_aprox_local_currency price_aprox_usd surface_total_in_m2 surface_covered_in_m2 price_per_m2 properati_url Country City Borough
    2 sell apartment |México|Distrito Federal|Cuauhtémoc| 19.41501,-99.175174 2700000.0 MXN 2748947.10 146154.51 61.0 61.0 44262.295082 http://cuauhtemoc.properati.com.mx/2pu_venta_a... México Distrito Federal Cuauhtémoc
    3 sell apartment |México|Distrito Federal|Cuauhtémoc| 19.41501,-99.175174 6347000.0 MXN 6462061.92 343571.36 176.0 128.0 49585.937500 http://cuauhtemoc.properati.com.mx/2pv_venta_a... México Distrito Federal Cuauhtémoc
    6 sell apartment |México|Distrito Federal|Miguel Hidalgo| 19.456564,-99.191724 670000.0 MXN 682146.11 36267.97 65.0 65.0 10307.692308 http://miguel-hidalgo-df.properati.com.mx/46h_... México Distrito Federal Miguel Hidalgo
    7 sell apartment |México|Distrito Federal|Gustavo A. Madero| 19.512787,-99.141393 1400000.0 MXN 1425379.97 75783.82 82.0 70.0 20000.000000 http://gustavo-a-madero.properati.com.mx/46p_v... México Distrito Federal Gustavo A. Madero
    8 sell house |México|Distrito Federal|Álvaro Obregón| 19.358776,-99.213557 6680000.0 MXN 6801098.67 361597.08 346.0 346.0 19306.358382 http://alvaro-obregon.properati.com.mx/46t_ven... México Distrito Federal Álvaro Obregón
    Let's check the example of a bar plot showing the value counts of each borough in the dataset. We first need to create a DataFrame showing the value counts:

    Let's check the example of a bar plot showing the value counts of each borough in the dataset. We first need to create a DataFrame showing the value counts:

    [5]:
    bar_df = pd.DataFrame(mexico_city1["Borough"].value_counts()).reset_index()
    [5]:
    index Borough
    0 Miguel Hidalgo 345
    1 Cuajimalpa de Morelos 255
    2 Álvaro Obregón 203
    3 Benito Juárez 198
    4 Tlalpan 171
    5 Iztapalapa 134
    6 Tláhuac 125
    7 Cuauhtémoc 120
    8 Gustavo A. Madero 89
    9 Venustiano Carranza 81
    10 Coyoacán 80
    11 La Magdalena Contreras 41
    12 Xochimilco 34
    13 Iztacalco 27
    14 Azcapotzalco 24
    15 Milpa Alta 1
    Since there are 16 different categories in Borough, we should increase the default plot size and rotate the x axis to make the plot more readable using the following syntax:

    Since there are 16 different categories in Borough, we should increase the default plot size and rotate the x axis to make the plot more readable using the following syntax:

    [6]:
    ax = sns.barplot(data=bar_df, x="index", y="Borough")
    [6]:
    [Text(0, 0, 'Miguel Hidalgo'),
     Text(1, 0, 'Cuajimalpa de Morelos'),
     Text(2, 0, 'Álvaro Obregón'),
     Text(3, 0, 'Benito Juárez'),
     Text(4, 0, 'Tlalpan'),
     Text(5, 0, 'Iztapalapa'),
     Text(6, 0, 'Tláhuac'),
     Text(7, 0, 'Cuauhtémoc'),
     Text(8, 0, 'Gustavo A. Madero'),
     Text(9, 0, 'Venustiano Carranza'),
     Text(10, 0, 'Coyoacán'),
     Text(11, 0, 'La Magdalena Contreras'),
     Text(12, 0, 'Xochimilco'),
     Text(13, 0, 'Iztacalco'),
     Text(14, 0, 'Azcapotzalco'),
     Text(15, 0, 'Milpa Alta')]
    <font size="+1">Practice</font>

    Practice

    Plot a bar plot showing the value counts for property types in "mexico-city-real-estate-1.csv":

    [7]:
    pro_typ_df = pd.DataFrame(mexico_city1["property_type"].value_counts()).reset_index()
    [7]:
    <AxesSubplot:xlabel='index', ylabel='property_type'>
    # Correlation Heatmaps

    Correlation Heatmaps¶

    A correlation heatmap shows the relative strength of correlations between the variables in a dataset. Here's what the code looks like:

    [8]:
    mexico_city1_numeric = mexico_city1.select_dtypes(include="number")
    [8]:
    <AxesSubplot:>
    Notice that we dropped the columns and rows with missing entries before plotting the graph.

    Notice that we dropped the columns and rows with missing entries before plotting the graph.

    This heatmap is showing us what we might already have suspected: the price is moderately positively correlated with the size of the properties.

    <font size="+1">Practice</font>

    Practice

    The seaborn documentation on heat maps indicates how to add numeric labels to each cell and how to use a different colormap. Modify the plot to use the viridis colormap, have a linewidth of 0.5 between each cell and have numeric labels for each cell.

    [ ]:
    ​x
     
    # References and Further Reading

    References and Further Reading¶

    • Official Plotly Express Documentation on Scatter Plots
    • Official Plotly Express Documentation on 3D Plots
    • Official Plotly Documentation on Notebooks
    • Plotly Community Forum Post on Axis Labeling
    • Plotly Express Official Documentation on Tile Maps
    • Plotly Express Official Documentation on Figure Display
    • Online Tutorial on String Conversion in Pandas
    • Official Pandas Documentation on using Lambda Functions on a Column
    • Official seaborn Documentation on Generating a Heatmap
    • Online Tutorial on Correlation Matrices in Pandas
    • Official Pandas Documentation on Correlation Matrices
    • Official Matplotlib Documentation on Colormaps
    • Official Pandas Documentation on Box Plots
    • Online Tutorial on Box Plots
    • Online Tutorial on Axes Labels in seaborn and Matplotlib
    • Matplotlib Gallery Example of an Annotated Heatmap
    ---

    Copyright 2022 WorldQuant University. This content is licensed solely for personal use. Redistribution or publication of this material is strictly prohibited. WQU WorldQuant University Applied Data Science Lab QQQQ

    ---

    Copyright 2022 WorldQuant University. This content is licensed solely for personal use. Redistribution or publication of this material is strictly prohibited.

    • 08-visualization-plotly.ipynb
    • 09-visualization-seaborn.ipynb
    • 10-databases-sql.ipynb
    • 11-databases-mongodb.ipynb
    • 12-ml-core.ipynb
    • 13-ml-data-pre-processing-and-production.ipynb
    • 14-ml-classification.ipynb
    • 15-ml-regression.ipynb
    xxxxxxxxxx
    ---
    Advanced Tools
    xxxxxxxxxx
    xxxxxxxxxx

    -

    Variables

    Callstack

      Breakpoints

      Source

      xxxxxxxxxx
      1
      15-ml-regression.ipynb
      • Linear Regression
      • Fitting a Model to Training Data
      • Generating Predictions Using a Trained Model
      • Ridge Regression
      • Generalization
      • Calculating the Mean Absolute Error for a List of Predictions
      • Access an Attribute of a Trained Model
      • Multicollinearity
        0
        8
        Python 3 (ipykernel) | Idle
        Saving completed
        Uploading…
        15-ml-regression.ipynb
        English (United States)
        Spaces: 4
        Ln 1, Col 1
        Mode: Command
        • Console
        • Change Kernel…
        • Clear Console Cells
        • Close and Shut Down…
        • Insert Line Break
        • Interrupt Kernel
        • New Console
        • Restart Kernel…
        • Run Cell (forced)
        • Run Cell (unforced)
        • Show All Kernel Activity
        • Debugger
        • Continue
          Continue
          F9
        • Evaluate Code
          Evaluate Code
        • Next
          Next
          F10
        • Step In
          Step In
          F11
        • Step Out
          Step Out
          Shift+F11
        • Terminate
          Terminate
          Shift+F9
        • Extension Manager
        • Enable Extension Manager
        • File Operations
        • Autosave Documents
        • Open from Path…
          Open from path
        • Reload Notebook from Disk
          Reload contents from disk
        • Revert Notebook to Checkpoint
          Revert contents to previous checkpoint
        • Save Notebook
          Save and create checkpoint
          Ctrl+S
        • Save Notebook As…
          Save with new path
          Ctrl+Shift+S
        • Show Active File in File Browser
        • Trust HTML File
        • Help
        • About JupyterLab
        • Jupyter Forum
        • Jupyter Reference
        • JupyterLab FAQ
        • JupyterLab Reference
        • Launch Classic Notebook
        • Licenses
        • Markdown Reference
        • Reset Application State
        • Image Viewer
        • Flip image horizontally
          H
        • Flip image vertically
          V
        • Invert Colors
          I
        • Reset Image
          0
        • Rotate Clockwise
          ]
        • Rotate Counterclockwise
          [
        • Zoom In
          =
        • Zoom Out
          -
        • Kernel Operations
        • Shut Down All Kernels…
        • Launcher
        • New Launcher
        • Main Area
        • Activate Next Tab
          Ctrl+Shift+]
        • Activate Next Tab Bar
          Ctrl+Shift+.
        • Activate Previous Tab
          Ctrl+Shift+[
        • Activate Previous Tab Bar
          Ctrl+Shift+,
        • Activate Previously Used Tab
          Ctrl+Shift+'
        • Close All Other Tabs
        • Close All Tabs
        • Close Tab
          Alt+W
        • Close Tabs to Right
        • Find Next
          Ctrl+G
        • Find Previous
          Ctrl+Shift+G
        • Find…
          Ctrl+F
        • Log Out
          Log out of JupyterLab
        • Presentation Mode
        • Show Header Above Content
        • Show Left Sidebar
          Ctrl+B
        • Show Log Console
        • Show Right Sidebar
        • Show Status Bar
        • Shut Down
          Shut down JupyterLab
        • Simple Interface
          Ctrl+Shift+D
        • Notebook Cell Operations
        • Change to Code Cell Type
          Y
        • Change to Heading 1
          1
        • Change to Heading 2
          2
        • Change to Heading 3
          3
        • Change to Heading 4
          4
        • Change to Heading 5
          5
        • Change to Heading 6
          6
        • Change to Markdown Cell Type
          M
        • Change to Raw Cell Type
          R
        • Clear Outputs
        • Collapse All Code
        • Collapse All Outputs
        • Collapse Selected Code
        • Collapse Selected Outputs
        • Copy Cells
          C
        • Cut Cells
          X
        • Delete Cells
          D, D
        • Disable Scrolling for Outputs
        • Enable Scrolling for Outputs
        • Expand All Code
        • Expand All Outputs
        • Expand Selected Code
        • Expand Selected Outputs
        • Extend Selection Above
          Shift+K
        • Extend Selection Below
          Shift+J
        • Extend Selection to Bottom
          Shift+End
        • Extend Selection to Top
          Shift+Home
        • Insert Cell Above
          A
        • Insert Cell Below
          B
        • Merge Cell Above
          Ctrl+Backspace
        • Merge Cell Below
          Ctrl+Shift+M
        • Merge Selected Cells
          Shift+M
        • Move Cells Down
        • Move Cells Up
        • Paste Cells Above
        • Paste Cells and Replace
        • Paste Cells Below
          V
        • Redo Cell Operation
          Shift+Z
        • Run Selected Cells
          Shift+Enter
        • Run Selected Cells and Don't Advance
          Ctrl+Enter
        • Run Selected Cells and Insert Below
          Alt+Enter
        • Run Selected Text or Current Line in Console
        • Select Cell Above
          K
        • Select Cell Below
          J
        • Split Cell
          Ctrl+Shift+-
        • Undo Cell Operation
          Z
        • Notebook Operations
        • Change Kernel…
        • Clear All Outputs
        • Close and Shut Down
        • Collapse All Cells
        • Deselect All Cells
        • Enter Command Mode
          Ctrl+M
        • Enter Edit Mode
          Enter
        • Expand All Headings
        • Interrupt Kernel
        • New Console for Notebook
        • New Notebook
          Create a new notebook
        • Reconnect To Kernel
        • Render All Markdown Cells
        • Restart Kernel and Clear All Outputs…
        • Restart Kernel and Run All Cells…
        • Restart Kernel and Run up to Selected Cell…
        • Restart Kernel…
        • Run All Above Selected Cell
        • Run All Cells
        • Run Selected Cell and All Below
        • Select All Cells
          Ctrl+A
        • Toggle All Line Numbers
          Shift+L
        • Toggle Collapse Notebook Heading
          T
        • Trust Notebook
        • Settings
        • Advanced Settings Editor
          Ctrl+,
        • Show Contextual Help
        • Show Contextual Help
          Live updating code documentation from the active kernel
          Ctrl+I
        • Spell Checker
        • Choose spellchecker language
        • Toggle spellchecker
        • Terminal
        • Decrease Terminal Font Size
        • Increase Terminal Font Size
        • New Terminal
          Start a new terminal session
        • Refresh Terminal
          Refresh the current terminal session
        • Use Terminal Theme: Dark
          Set the terminal theme
        • Use Terminal Theme: Inherit
          Set the terminal theme
        • Use Terminal Theme: Light
          Set the terminal theme
        • Text Editor
        • Decrease Font Size
        • Increase Font Size
        • Indent with Tab
        • New Markdown File
          Create a new markdown file
        • New Python File
          Create a new Python file
        • New Text File
          Create a new text file
        • Spaces: 1
        • Spaces: 2
        • Spaces: 4
        • Spaces: 8
        • Theme
        • Decrease Code Font Size
        • Decrease Content Font Size
        • Decrease UI Font Size
        • Increase Code Font Size
        • Increase Content Font Size
        • Increase UI Font Size
        • Theme Scrollbars
        • Use Theme: JupyterLab Dark
        • Use Theme: JupyterLab Light